System and method for mining data

Info

Publication number: 20030172053
Type: Application
Filed: Feb 3, 2003
Publication Date: Sep 11, 2003
Inventor: John Fairweather (Santa Monica, CA)
Application Number: 10357290

Abstract

A system and method for extracting data, hereinafter referred to as MitoMine™, that produces a strongly-typed ontology defined collection referencing (and cross referencing) all extracted records. The input to the mining process can be any data source, such as a text file delimited into a set of possibly dissimilar records. Mitomine contains parser routines and post-processing functions, known as ‘munchers’. The parser routines can be accessed either via a batch mining process or as part of a running server process connected to a live source. Munchers can be registered on a per data-source basis in order to process the records produced, possibly writing them to an external database and/or a set of servers. The present invention also embeds an interpreted ontology based language within a compiler/interpreter (for the source format) such that the statements of the embedded language are executed as a result of the source compiler ‘recognizing’ a given construct within the source and extracting the corresponding source content. In this way, the execution of the statements in the embedded program will occur in a sequence that is dictated wholly by the source content. This system and method therefore make it possible to bulk extract free-form data from such sources as CD-ROMs, the web etc. and have the resultant structured data loaded into an ontology based system.

Description

Description

BACKGROUND OF THE INVENTION

[0001] The data ingestion and conversion process is generally known as data mining, and the creation of robust systems to handle this problem is the subject of much research, and has spawned the creation of many specialized languages (e.g., Perl) intended to make this process easier. Unfortunately, while there have been some advances, the truth of the matter is that none of these ‘mining’ languages really provides anything more than a string manipulation library embedded into the language syntax itself. In other words, such languages are nothing more than shorthand for the equivalent operations written as a series of calls to a powerful subroutine library. A prerequisite for any complex data processing application, specifically a system capable of processing and analyzing disparate data sources, is a system that can convert the structured, semi-structured, and un-structured information sources into their equivalent representation in the target ontology, thereby unifying all sources and allowing cross-source analysis.

[0002] For example, in a current generation data-extraction script, the code involved in the extraction basically works its way through the text from beginning to end trying to recognize delimiting tokens and once having done so to extract any text within the delimiters and then assign it to the output data structure. When there is a one-to-one match between source data and target representation, this is a simple and effective strategy. As we widen the gap between the two, however, such as by introducing multiple inconsistent sources, increasing the complexity of the source, nesting information in the source to multiple levels, cross referencing arbitrarily to other items within the source, and distributing and interspersing the information necessary to determine an output item within a source, the situation rapidly becomes completely unmanageable by this technique, and highly vulnerable to the slightest change in source format or target data model. This mismatch is at the heart of all problems involving the need for multiple different systems to intercommunicate meaningful information, and makes conventional attempts to mine such information prohibitively expensive to create and maintain. Unfortunately for conventional mining techniques, much of the most valuable information that might be used to create truly intelligent systems comes from publishers of various types. Publishing houses make their money from the information that they aggregate, and thus are not in the least bit interested in making such information available in a form that is susceptible to standard data mining techniques. Furthermore, most publishers deliberately introduce inconsistencies and errors into their data in order both to detect intellectual property rights violations by others, and to make automated extraction as difficult as possible. Each publisher, and indeed each title from any given publisher, uses different formats, and has an arrangement that is custom tailored to the needs of whatever the publication is. The result is that we are faced with a variety of source formats on CD-ROMs, databases, web sites, and other legacy systems that completely stymie standard techniques for acquisition and integration. Very few truly useful sources are available in a nice neat tagged form such as XML and thus to rely on markup languages such as XML to aid in data extraction is a woefully inadequate approach in real-world situations.

[0003] One of the basic problems that makes the extraction process difficult is that the control-flow based program that is doing the extraction has no connection to the data itself (which is simply input) and must therefore invest huge amounts of effort extracting and keeping track of its ‘state’ in order to know what it should do with information at any given time. What is needed, then, is a system in which the content of the data itself actually determines the order of execution of statements in the mining language and automatically keeps track of the current state. In such a system, whenever an action was required of the extraction code, the data would ‘tell’ it to take that action, and all of the complexity would melt away. Assuming such a system is further tied to a target system ontology, the mining problem would become quite simple. Ideally, such a solution would tie the mining process to compiler theory, since that is most powerful formalized framework available for mapping source textual content into defined actions and state in a rigorous and extensible manner. It would also be desirable to have an interpreted language that is tied to the target ontology (totally different from the source format), and for which the order of statement execution could be driven by source data content

SUMMARY OF INVENTION

[0004] The system of this invention takes the data mining process to a whole new level of power and versatility by recognizing that, at the core of our past failings in this area, lies the fact that conventional control-flow based programming languages are simply not suited to the desired system, and must be replaced at the fundamental level a more flexible approach to software system generation. There are two important characteristics of the present invention that help create this paradigm shift. The first is that, in the preferred embodiment, the system of the present invention includes a system ontology such that the types and fields of the ontology can be directly manipulated and assigned within the language without the need for explicit declarations. For example, to assign a value to a field called “notes.sourceNotes” of a type, the present invention would only require the statement “notes.sourceNotes=”. An ontology is an explicit formal specification of how to represent the objects, concepts and other entities that are assumed to exist in some area of interest and the relationships that hold among them. The second, and one of the most fundamental characteristics, is that the present invention gives up on the idea of a control-flow based programming language (i.e., one where the order of execution of statements is determined by the order of those statements within the program) in order to dramatically simplify the extraction of data from a source. In other words, the present invention represents a radical departure from all existing “control” notions in programming.

[0005] The present invention, hereinafter referred to as MitoMine™, is a generic data extraction capability that produces a strongly-typed ontology defined collection referencing (and cross referencing) all extracted records. The input to the mining process tends to be some form of text file delimited into a set of possibly dissimilar records. Mitomine contains parser routines and post processing functions, known as ‘munchers’. The parser routines can be accessed either via a batch mining process or as part of a running server process connected to a live source. Munchers can be registered on a per data-source basis in order to process the records produced, possibly writing them to an external database and/or a set of servers. The present invention embeds an interpreted ontology based language within a compiler/interpreter (for the source format) such that the statements of the embedded language are executed as a result of the source compiler ‘recognizing’ a given construct within the source and extracting the corresponding source content. In this way, the execution of the statements in the embedded program will occur in a sequence that is dictated wholly by the source content. This system and method therefore make it possible to bulk extract free-form data from such sources as CD-ROMs, the web etc. and have the resultant structured data loaded into an ontology based system.

[0006] In the preferred embodiment, a MitoMine™ parser is defined using three basic types of information:

[0007] 1) A named source-specific lexical analyzer specification

[0008] 2) A named BNF specification for parsing the source

[0009] 3) A set of predefined plug-in functions capable of interpreting the source information via C** statements.

[0010] Other improvements and extentions to this system will be defined herein.

BRIEF DESCRIPTION OF THE FIGURES

[0011] [NONE]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

[0012] The present invention is built upon this and, in the preferred embodiment, uses a number of other key technologies and concepts. For example, these following patent applications (which are expressly incorporated herein) disclose all the components necessary to build up a system capable of auto-generating all user interface, storage tables, and querying behaviors required in order to create a system directly from the specifications given in an ontology description language (ODL). These various building-block technologies have been previously described in the following patent applications:

[0013] 1) Appendix 1—Memory Patent

[0014] 2) Appendix 2—Lexical Patent

[0015] 3) Appendix 3—Parser Patent

[0016] 4) Appendix 4—Types Patent

[0017] 5) Appendix 5—Collections Patent

[0018] 6) Appendix 6—Ontology Patent

[0019] In the Parser Patent, a system was described that permits execution of the statements in the embedded program in a sequence that is dictated wholly by the source content, in that the ‘reverse polish’ operators within that system are executed as the source parse reaches an appropriate state and, as further described in that patent, these operators are passed a plug-in hint string when invoked. In the preferred embodiment, the plug-in hint string will be the source for the interpreted ontology-based language and the plug-ins themselves will invoke an inner level parser in order to execute these statements. The Ontology Patent introduced an ontology based language that is an extension of the C language known as C*. This is the preferred ontology based language for the present invention. We will refer to the embedded form of this language as C**, the extra ‘*’ symbol being intended to imply the additional level of indirection created by embedding the language within a source format interpreter. The output of a mining process will be a set of ontology defined types (see Types Patent) within a flat data-model collection (see Memory Patent and Collection Patent) suitable for instantiation to persistent storage and subsequent query and access via the ontology (see patent reference 6).

[0020] In the preferred embodiment, a MitoMine™ parser is defined using three basic types of information:

[0021] 1) A named source-specific lexical analyzer specification

[0022] 2) A named BNF specification for parsing the source

[0023] 3) A set of predefined plug-in functions capable of interpreting the source information via C** statements.

[0024] The BNF format may be based upon any number of different BNF specifications. MitoMine™ provides the following additional built-in parser plug-ins which greatly facilitate the process of extracting unstructured data into run-time type manager records:

<@1:1>

<@1:2>

[0025] These two plug-ins delimit the start and end of an arbitrary possibly multi-lined string to be assigned to the field designated by the following call to <@1:5:fieldPath=$>. This is the method used to extract large arbitrary text fields. The token sequence for these plug-ins is always of the form <@1:1><1:String><@1:2>, that is any text occurring after the appearance of the <@1:1> plug-in on the top of the parsing stack will be converted into a single string token (token #1) which will be assigned on the next <@1:5> plug-in. The arbitrary text will be terminated by the occurrence of any terminal in the language (defined in the LEX specification) whose value is above 128. Thus the following snippet of BNF will cause the field ‘pubName’ to be assigned whatever text occurs between the token <PUBLICATION> and <VOLUME/ISSUE> in the input file:

[0026] In the preferred embodiment, when extracting these arbitrary text fields, all trailing and leading white space is removed from the string before assignment, and all occurrences of LINE_FEED are removed to yield a valid text string. The fact that tokens below 128 will not terminate the arbitrary text sequence is important in certain situations where a particular string is a terminal in the language and yet might also occur within such a text sequence where it should not be considered to have any special significance. All such tokens can be assigned token numbers below 128 in the LEX specification thus ensuring that no confusion arises. The occurrence of another <@1:1> or a <@1:4> plug-in causes any previous <1:String> text accumulated to be discarded. A <@1:5> causes execution of a C** statements that generally cause extracted information to be assigned to the specified field and then clears the record of the accumulation. If a plug-in hint consisting of a decimal number follows the <@1:1> as in <@1:1:4> that number specifies the maximum number of lines of input that will be consumed by the plug-in (four in this example). This is a useful means to handle input where the line number or count is significant.

<@1:3>

[0027] In the preferred embodiment, the occurrence of this plug-in indicates that the extraction of a particular record initiated by the <@1:4> plug-in is complete and should be added to the collection of records extracted.

<@1:4:typeName>

[0028] In the preferred embodiment, the occurrence of the plug-in above indicates that the extraction of a new record of the type specified by the ‘typeName’ string is to begin. The “typename” will preferably match a known type manager type either defined elsewhere or within the additionally type definitions supplied as part of the parser specification.

<@1:5:C** assignment(s)>

[0029] In the preferred embodiment, the plug-in above is used to assign values to either a field or a register. Within the assigned expression, the previously extracted field value may be referred to as ‘$’. Fields may be expressed as a path to sub-fields of the structure to any depth using normal type manager path notation (same as for C). As an example, the field specifier “description[$aa].u.equip.specifications” refers to a field within the parent structure that is within an array of unions. The symbol ‘$aa’ is a register designator. There are 26*26 registers ‘$aa’ to ‘$zz’ which may be used to hold the results of calculations necessary to compute field values. A single character register designator may also be used instead thus ‘$a’ is the same as ‘$aa’, ‘$b’ is the same as ‘$ba’ etc. Register names may optionally be followed by a text string (no spaces) in order to improve readability (as in $aa:myIndex) but this text string is ignored by the C** interpreter. The use of registers to store extracted information and context is key to handling the distributed nature of information in published sources. In the example above, ‘$a’ is being used as an index into the array of ‘description’ fields. To increment this index a “<@1:5:$a=$a+1>” plug-in call would be inserted in the appropriate part of the BNF (presumably after extraction of an entire ‘description’ element). All registers are initially set to zero (integer) when the parse begins, thereafter their value is entirely determined by the <@1:5> plug-ins that occur during the extraction process. If a register is assigned a real or string value, it adopts that type automatically until a value of another type is assigned to it. Expressions may include calls to functions (of the form $FuncName), which provide a convenient means of processing the inputs extracted into certain data types for assignment. These functions provide capabilities comparable to the string processing libraries commonly found with older generation data mining capabilities.

[0030] When assigning values to fields, the <@1:5> plug-in performs intelligent type conversions, for example:

[0031] 1) If the token is a <1:String> and the field is a ‘charHdI’, a handle is created and assigned to the field. Similarly for a ‘charPtr’. If the field is a fixed length character array, the string is copied into it. If it won't fit, a bounds error is flagged. If the field is already non-empty (regardless of type) then the <@1:5> plugin appends any new text to the end of the field value (if possible). Note that registers do not append automatically unless you use the syntax $a=$a+“string”.

[0032] 2) If the field is numeric, appropriate type conversions from the extracted value occur. Range checking could be automatic. Multiple assignments may be separated by semi-colons. The full syntax supported within the ‘assignment’ string is defined by the system BNF language “MitoMine” (described below).

[0033] Note that because the order of commutative operator (e.g., “+”) evaluation is guaranteed to be left-to-right, multiple non-parenthesized string concatenation operations can be safely expressed as a single statement as in:

fieldname=“Hello”+$FirstCapOnly($a)+“do you like”+$b+“\n”

[0034] The <@1:5> plug-in may also be used to support limited conditional statements which may be performed using the ‘if’ and ‘ifelse’ keywords. The effect of the ‘if’ is to conditionally skip the next element of the production that immediately follows the <@1:5> containing the ‘if’ (there can be only one statement within an ‘if’ or ‘ifelse’ block). For example:

<@1:5:if(1==0)><@1:4:typeName>

[0035] would cause the <@1:4> plug-in to be discarded without interpretation. Similarly:

<@1:5:ifelse(1==0)><@1:4:typeName1><@1:4:typeName2>

[0036] causes execution of the second of the two <@1:4> plug-ins while:

<@1:5:ifelse(0= =)><@1:5:$a=$a+1;$b=1><@1:5:$a=$a−1;$b=0>

[0037] causes execution of the first block to increment $a and assign $b to 1.

[0038] More significantly, since it is possible to discard any element from the production in this manner, the prudent use of conditional <@1:5> evaluation can be used to modify the recognized syntax of the language. Consider the following production:

myProduction::=<@1:5:ifelse($a>=0)>positive_prod negative_prod

[0039] In this example, the contents of register ‘$a’ is determining which of two possible productions will get evaluated next. This can be a very powerful tool for solving non-context-free language ambiguities (normally intractable to this kind of parser) by remembering the context in one of the registers and then resolving the problem later when it occurs. The results of misusing this capability can be very confusing and the reader is referred to the incorporated materials of the Parser Patent for additional details. That having been said, the following simplified guidelines should help to ensure correctness:

[0040] For any production of the form:

prod::=<@1:5:ifelse (expression)> thenClause elseClause

[0041] Ensure:

[0042] 1) FIRST(thenClause)==FIRST(elseClause)

[0043] 2) Either both thenClause and elseClause are NULLABLE, or neither is

[0044] 3) If elseClause is not NULLABLE, and if necessary (depending on other occurences of thenClause),

[0045] include a production elsewhere {that may never be executed} to ensure that FOLLOW(thenClause) includes FOLLOW(elseClause)

[0046] For any production of the form:

prod::=prevElement<@1:5:if(expression)>thenClause nextElement

[0047] Ensure that if thenClause is not NULLABLE, and if necessary (depending on other occurences of nextElement), include a production elsewhere {that may never be executed} to ensure that FIRST(nextElement) is entirely contained within FOLLOW(prevElement).

[0048] Note that all plug-ins may contain multiple lines of text by use of the <cont> symbol (see Parser patent). This may be required in the case where a <@1:5> statement exceeds the space available on a single line (e.g., many parameters to a function). The maximum size of any given plug-in text in the preferred embodiment is 8 KB.

[0049] The present invention also permits the specification of the language specific parser to include any user dialogs and warnings that might be required for the parser concerned, any additional type definitions that might be required as part of parser operation, and any custom annotations and scripts (see Collections Patent) that might be necessary.

[0050] Within the <@1:5> plug-in, in addition to supporting conditionals, additive, multiplicative and assignment operators, this package preferably provides a number of built-in functions that may be useful in manipulating extracted values in order to convert them to a form suitable for assignment to typed fields. These functions are loosely equivalent to the string processing library of conventional mining languages. Function handlers may be registered (via a registry API—see Parser Patent for further details) to provide additional built in functions. In the built-in function descriptions below, for example, the type of a given parameter is indicated between square brackets. The meaning of these symbols in this example is as follows:

[0051] [I]—Integer value (64 bit)

[0052] [F]—Floating point value (double)

[0053] [S]—String value

[0054] The following is a partial list of predefined built-in functions that have been found to be useful in different data mining situations. New functions may be added to this list and it is expected that use of the system will often include the step of adding new functions. In such a case, if a feature is not provided, it can be implemented and registered as part of any particular parser definition. On the other hand, none of the features listed below are required meaning that a much smaller set of functions could also be used. In the preferred embodiment, however, the following functions (or ones having similar functionality) would be available.

[0055] 1) [F] $Date( )

[0056] get current date/time into a date-double

[0057] 2) [F] $StringToDate([S] dateString,[S] calendar)

[0058] convert “dateString” to date/time double, current date if date string format invalid. The currently supported calendar values are “G”—Gregorian, “J”—Julian etc. Note that in the Gregorian calendar you may specify the date string in a wide variety of formats, in any other calendar it must be in the following format: “yyyy:mm:dd [hh:mm[:ss] [AM/PM]]”

[0059] 3) [S] $TextAfter([S] srcStr,[S] delimStr)

[0060] Return the string portion after the specified delimiter sequence. Returns“” if not found.

[0061] 4) [S] $TextBefore([S] srcStr,[S] delimStr)

[0062] Return the string portion before the specified delimiter sequence. Returns ″″ if not found.

[0063] 5) [S] $TextBetween([S] srcStr,[S] startStr,[S] endStr)

[0064] Return the string portion between the specified delimiter sequences. Returns“” if not found.

[0065] 6) [I] $Integer([S] aString)

[0066] Convert the specified string to an integer (decimal or hex)

[0067] 7) [F] $Real([S] aString)

[0068] Convert the specified string to a real number

[0069] 6) [I] $IntegerWithin([S] aString,[I] n)

[0070] Extract the n'th integer (decimal or hex, n=1 . . . ) within the specified arbitrary string

[0071] 7) [F] $RealWithin([S] aString,[I] n)

[0072] Extract the n'th real (n=1 . . . ) within the specified arbitrary string

[0073] 8) [S] $StripMarkup([S] aString)

[0074] Strip any Markup language tags out of a string to yield plain text.

[0075] 9) [S] $SourceName( )

[0076] Inserts the current value of ‘languageName’

[0077] 10) [S] $SetPersRefInfo([S] aString)

[0078] This function allows you to append to the contents of the ‘stringH’ field of a persistent reference field rather than assigning to the name. The function result is equal to ‘aString’ but the next assignment made by the parser will be to the ‘stringH’ sub-field, not the ‘name’ sub-field.

[0079] 11) [S] $FirstCapOnly([S] aString)

[0080] Converts a series of words in upper/lower case such that each word starts with an upper case character and all subsequent characters are lower case.

[0081] 12) [S] $TextNotAfter([S] srcStr,[S] delimStr)

[0082] Similar in operation to $TextBefore( ) except if ‘delimStr’ is not found, the original string is returned un-altered.

[0083] 13) [S] $TextNotBefore([S] srcStr,[S] delimStr)

[0084] Similar in operation to $TextAfter( ) except if ‘delimStr’ is not found, the original string is returned un-altered.

[0085] 14) [S] $TextNotBetween([S] srcStr,[S] startStr,[S] endStr)

[0086] Returns what remains after removing the string portion between the specified delimiter sequences (and the delimiter sequences themselves). If the sequence is not found, the original string is returned un-altered.

[0087] 15) [S] $TruncateText([S] srcStr,[I] numChars)

[0088] Truncated the source string to the specified number of characters.

[0089] 16) [S] $TextBeforeNumber([S] srcStr)

[0090] This function is similar in operation to $TextBefore( ) but the ‘delimStr’ is taken to be the first numeric digit encountered.

[0091] 17) [S] $TextWithout([S] srcStr,[S] sequence)

[0092] This function removes all occurrences of the specified sequence from the source string.

[0093] 18) [S] $WordNumber([S] srcStr,[I] number)

[0094] This function gets the specified word (starting from 1) from the source string. If ‘number’ is negative, the function counts backward from the last word in the source string.

[0095] 19) [S] $Ask([S] promptStr)

[0096] This function prompts the user using the specified string and allows him to enter a textual response which is returned as the function result.

[0097] 20) [S] $TextWithoutBlock([S] srcStr,[S] startDelim,[S] endDelim)

[0098] This function removes all occurences of the delimited text block (including delimiters) from the source string.

[0099] 21) [S] $ReplaceSequence([S] srcStr,[S] sequence,[S] nuSequence)

[0100] This function replaces all occurences of the target sequence by the sequence ‘nuSequence’ within the given string.

[0101] 22) [S] $AppendIfNotPresent([S] srcStr,[S] endDelim)

[0102] This function determines if ‘srcStr’ ends in ‘endDelim’ and if not appends ‘endDelim’ to ‘srcStr’ returning the result.

[0103] 23) [S] $ProperNameFilter([S] srcStr,[I] wordMax,[S] delim)

[0104] This function performs the following processing (in order) designed to facilitate the removal of extraneous strings of text from ‘delim’ separated lists of proper names (i.e., capitalized first letter words):

[0105] a) if the first non-white character in a ‘delim’ bounded block is not upper case, remove the entire string up to and including the trailing occurence of ‘delim’ (or end of string).

[0106] b) for any ‘delim’ bounded block, strip off all trailing words that start with lower case letters.

[0107] c) if more than ‘wordMax’ words beginning with a lower case letter occur consecutively between two occurrences of ‘delim’, terminate the block at the point where the consecutive words occur.

[0108] 24) [S] $Sprintf([S] formatStr, . . . )

[0109] This function performs a C language sprintf( ) function, returning the generated string as its result.

[0110] 25) [S] $ShiftChars([S] srcStr,[I] delta)

[0111] This function shifts the character encoding of all elements of ‘srcStr’ by the amount designated in ‘delta’ returning the shifted string as a result. This functionality can be useful for example when converting between upper and lower case.

[0112] 26) [S] $FlipChars([S] srcStr)

[0113] This function reverses the order of all characters in ‘srcStr’.

[0114] 27) [S] $ReplaceBlockDelims([S] srcStr,[S] startDelim,[S] endDelim,[S] nuStartDelim,[S] nuEndDelim,[I] occurence,[I] reverse)

[0115] This function replaces the start and end delimiters of one or more delimited blocks of text by the new delimiters specified. If ‘occurence’ is zero, all blocks found are processed, otherwise just the block specified (starting from 1). If ‘reverse’ is non-zero (i.e.,1), this function first locates the ending delimiter and then works backwards looking for the start delimiter. Often if the start delimiter is something common like a space character (e.g., looking for the last word of a sentence), the results of this may be quite different from those obtained using ‘reverse’=0.

[0116] 28) [S] $RemoveIfFollows([S] srcStr,[S] endDelim)

[0117] This function determines if ‘srcStr’ ends in ‘endDelim’ and if so removes ‘endDelim’ from ‘srcStr’ returning the result.

[0118] 29) [S] $RemoveIfStarts([S] srcStr,[S] startDelim)

[0119] This function determines if ‘srcStr’ starts with ‘startDelim’ and if so removes ‘startDelim’ from ‘srcStr’ returning the result.

[0120] 30) [S] $PrependIfNotPresent([S] srcStr,[S] startDelim)

[0121] This function determines if ‘srcStr’ starts with ‘startDelim’ and if not prepends ‘startDelim’ to ‘srcStr’ returning the result.

[0122] 31) [S] $NoLowerCaseWords([S] srcStr)

[0123] This function eliminates all words beginning with lower case letters from ‘srcStr’ returning the result.

[0124] 32) [S] $ReplaceBlocks([S] srcStr,[S] startDelim,[S] endDelim,[I] occurence,[S] nuSequence)

[0125] This function replaces one or all blocks delimited by the specified delimiter sequences with the replacement sequence specified. If ‘occurrence’ is zero, all blocks are replaced, otherwise the occurrence is a one-based index to the block to replace.

[0126] 33) [S] $AppendIfNotFollows([S] srcStr,[S] endDelim)

[0127] This function determines if ‘srcStr’ ends in ‘endDelim’ and if not appends ‘endDelim’ to ‘srcStr’ returning the result.

[0128] 34) [I] $WordCount([S] srcStr)

[0129] This function counts the number of words in the source string, returning the numeric result.

[0130] 35) [S] $PreserveParagraphs([S] srcStr)

[0131] This function eliminates all line termination characters (replacing them by spaces) in the source string other than those that represent paragraph breaks. Source text has often been formatted to fit into a fixed page width (e.g., 80 characters) and since we wish the captured text to re-size to fit whatever display area is used, it is often necessary to eliminate the explicit line formatting from large chunks of text using this function. A paragraph is identified by a line termination immediately followed by a tab or space character (also works with spaces for right justified scripts), all other explicit line formatting is eliminated. The resulting string is returned.

[0132] 36) [I] $StringSetIndex([S] srcStr,[I] ignoreCase,[S] setStr1 . . . [S] setStrN)

[0133] This function compares ‘srcStr’ to each of the elements in the set of possible match strings supplied, returning the index (starting from 1) of the match string found, or zero if no match is found. If ‘ignoreCase’ is non-zero, the comparisons are case insensitive, otherwise they are exact.

[0134] 37)[S] $IndexStringSet([I] index,[S] setStr1 . . . [S] setStrN)

[0135] This function selects a specific string from a given set of strings by index (1-based), returning as a result the selected string. If the index specified is out of range, an empty string is returned.

[0136] 38) [S] $ReplaceChars([S] srcStr,[S] char,[S] nuChar)

[0137] This function replaces all occurrences of ‘char’ in the string by ‘nuChar’ returning the modified string as a result.

[0138] 39) [S] $Sentence([S] srcStr,[I] index)

[0139] This function extracts the designated sentence (indexing starts from 0) from the string, returning as a result the sentence. If the index specified is negative, the index counts backwards from the end (i.e., −1 is the last sentence etc.). A sentence is identified by any sequence of text terminated by a period.

[0140] 40) [S] $FindHyperlink([S] srcStr,[S] domain, [I] index)

[0141] This function will extract the index'th hyperlink in the hyperlink domain specified by ‘domain’ that exists in ‘srcStr’ (if any) and return as a result the extracted hyperlink name. This technique can be used to recognize known things (e.g., city or people names) in an arbitrary block of text. If no matching hyperlink is found, the function result will be an empty string.

[0142] 41) [S] $AssignRefType([S] aString)

[0143] This function allows you to assign directly to the typeID sub-field of a persistent reference field rather than assigning to the name. The function result is equal to ‘aString’ but the next assignment made by the parser will be to the typeID sub-field ‘aString’ is assumed to be a valid type name), not the ‘name’ sub-field.

[0144] 42) [I] $RecordCount( )

[0145] This function returns the number of records created so far during the current mining process.

[0146] 43) [S] $Exit([S] aReason)

[0147] Calling this function causes the current parsing run to exit cleanly, possibly displaying a reason for the exit (to the console) as specified in the ‘aReason’ string (NULL if no reason given).

[0148] 44) [I] $MaxRecords( )

[0149] This function returns the maximum number of records to be extracted for this run. This value can either be set by calling $SetMaxRecords( ) or it may be set by external code calling MN_SetMaxRecords( ).

[0150] 45) [I] $SetMaxRecords([I] max)

[0151] This function sets the maximum number of records to be extracted for this run. See $MaxRecords( ) for details.

[0152] 46) [I] $FieldSize([S] fieldName)

[0153] This function returns the size in bytes of the field specified in the currently active type record as set by the preceeding <@1:4:typeName> operator. Remember that variable sized string fields (i.e., char @fieldName) and similar will return a size of sizeof(Ptr), not the size of the string within it.

[0154] 47) [I] $TextContains([S] srcText,[S] subString)

[0155] This function returns 0 if the ‘srcText’ does not contain ‘subString’, otherwise it returns the character index within ‘srctext’ where ‘subString’ starts+1.

[0156] 48) [I] $ZapRegisters([S] minReg,[S] maxReg)

[0157] This function empties the contents of all registers starting from ‘minReg’ and ending on ‘maxReg’. The parameters are simply the string equivalent of the register name (e.g., “$aa”). When processing multiple records, the use of $ZapRegisters( ) is often more convenient than explicit register assignments to ensure that all the desired registers start out empty as record processing begins. The result is the count of the number of non-empty registers that were zapped.

[0158] 49) [I] $CRCString([S] srcText)

[0159] This function performs a 32-bit CRC similar to ANSI X3.66 on the text string supplied, returning the integer CRC result. This is can be useful when you want to turn an arbitrary (i.e., non-alphanumeric) string into a form that is (probably!) unique for name generating or discriminating purposes.

[0160] Note that parameters to routines may be either constants (of integer, real or string type), field specifiers referring to fields within the current record being extracted, registers, $ (the currently extracted field value), or evaluated expressions which may include embedded calls to other functions (built-in or otherwise). This essentially creates a complete programming language for the extraction of data into typed structures and collections. The C** programming language provided by the <@1:5> plug-ins differs from a conventional programming language in that the order of execution of the statements is determined by the BNF for the language and the contents of the data file being parsed. In the preferred embodiment, the MitoMine™ parser is capable of recognizing and evaluating the following token types:

[0161] 3—DecInt—syntax as for a C strtoul( ) call but ignores embedded commas.

[0162] 4—Real—real—as for C strtod( )

[0163] 5—Real—real scientific format—as for C strtod( )

[0164] The plug-in 5 MitoMine™ parser, in addition to recognizing registers, $, $function names, and type field specifications, can also preferably recognize and assign the following token types:

[0165] 2—character constant (as for C)

[0166] 7—Hex integer (C format)

[0167] 3—decimal integer (as for C strtoul)

[0168] 10—octal integer (as for strtoul)

[0169] 4—real (as for strtod)

[0170] 5—real with exponent (as for strtod)

[0171] 12—string constant (as for C).

[0172] Character constants can be a maximum of 8 characters long, during input, they are not sign extended. The following custom parser options would preferably be supported:

[0173] kTraceAssignments (0x00010000)—Produces a trace of all <@1:5> assignments on the console

[0174] kpLineTrace (0x00020000)—Produces a line trace on the console

[0175] kTraceTokens (0x00040000)—Produces a trace of each token recognized

[0176] These options may be specified for a given parser language by adding the corresponding hex value to the parser options line. For example, the specification below would set kTraceAssignments+kpLineTrace options in addition to those supported by the basic parse package:

=0x30000+kPreserveBNFsymbols+kBeGreedyParser

[0177] The lexical analyzer options line can also be used to specify additional white-space and delimiter characters to the lexical analyzer as a comma separated list. For example the specification below would cause the characters ‘a’ and ‘b’ to be treated as whitespace (see LX_AddWhiteSpace) and the characters ‘Y’ and ‘Z’ to be treated as delimiters (see LX_AddDelimiter).

=kNoCaseStates+whitespace(a,b)+delimiter(Y,Z)

[0178] Appendix A provides a sample of the BNF and LEX specifications that define the syntax of the <@1:5> plug-in (i.e., C**) within MitoMine™ (see Parser Patent for further details). Note that most of the functionality of C** is already provided by the predefined plug-in functions (plug-in 0) supplied by the basic parser package. A sample implementation of the <@1:5> plug-in one and a sample implementation of a corresponding resolver function are also provided.

[0179] As described previously, the lexical and BNF specifications for the outermost parser vary depending on the source being processed (example given below), however the outer parser also has a single standard plug-in and resolver. A sample implementation of the standard plug-in one and a sample implementation of a corresponding resolver function are also provided in Appendix A.

[0180] The listing below gives the API interface to the MitoMine™ capability for the preferred embodiment although other forms are obviously possible. Appendix A provides the sample pseudo code for the API interface.

[0181] In the preferred embodiment, a function, hereinafter called MN_MakeParser( ), initializes an instance of the MitoMine™ and returns a handle to the parser database which is required by all subsequent calls. A ‘parserType’ parameter could be provided to select a particular parsing language to be loaded (see PS_LoadBNF) and used.

[0182] In the preferred embodiment, a function, hereinafter called MN_SetRecordAdder( ) determines how (or if) records once parsed are added to the collection. The default record adder creates a set of named lists where each list is named after the record type it contains.

[0183] In the preferred embodiment, a function, hereinafter called MN_SetMineFunc( ), sets the custom mine function handler for a MitoMine™ parser. Additional functions could also be defined over and above those provided by MitoMine™ within the <@1:5: . . . > plugin context. A sample mine function handler follows: 1 static Boolean myFunc ( // custom function handler ET_ParseHdl aParseDB, //IO:handle to parser DB int32 aContextID //I:context {cube root} ) // R:TRUE for success p = (myContextPtr)aContextID; // get our context pointer opCount = PS_GetOpCount(aParseDB,TOP); // get # of operands tokp = PS_GetToken(aParseDB,opCount); // get fn name for ( i = 0 ; i < opCount ; i++ ) if ( ?PS_EvalIdent(aParseDB, i) ) // eval all elements on stack { res = NO; goto BadExit; } if ( ?US_strcmp(tokp, ″$myFuncName″) ) // function name { -- check operand count and type -- implement function -- set resulting value into stack ′opCount′ e.g.: PS_SetiValue(aParseDB,opCount,result); else if ( ?US_strcmp(tokp, ″$another function″) )

[0184] In the preferred embodiment, a function, hereinafter called MN_SetMaxRecords( ), sets the maximum number of records to be mined for a MitoMine™ parser. This is the number returned by the built-in function $GetMaxRecords( ). If the maximum number of records is not set (i.e., is zero), all records are mined until the input file(s) is exhausted.

[0185] In the preferred embodiment, a function, hereinafter called MN_SetMineLineFn( ), sets the MitoMine™ line processing function for a given MitoMine™ parser. A typical line processing function might appear as follows: 2 static void myLineFn ( // Built-in debugging mine-line fn ET_ParseHdl aParseDB, // I:Parser DB int32 aContextID, // I:Context int32 lineNum, // I:Current line number charPtr lineBuff, // IO:Current line buffer charPtr aMineLineParam // I:String parameter to function ) // R:void

[0186] These functions can be used to perform all kinds of different useful functions such as altering the input stream before the parse sees it, adjusting parser debugging settings, etc. The ‘aMineLineParam’ parameter above is an arbitrary string and can be formatted any way you wish in order to transfer the necessary information to the line processing function. The current value of this parameter is set using MN_SetMineLineParam( ).

[0187] In the preferred embodiment, a function, hereinafter called MN_SetMineLineParam( ), sets the string parameter to a MitoMine™ line processing function.

[0188] In the preferred embodiment, two functions, hereinafter calleds MN_SetParseTypeDB( ) and MN_GetParseTypeDB( ), can be used to associate a type DB (probably obtained using MN_GetMineLanguageTypeDB) with a MitoMine™ parser. This is preferable so that the plug-ins associated with the extraction process can determine type information for the structures unique to the language. In the preferred embodiment, the function MN_GetParseTypeDB( ) would return the current setting of the parser type DB.

[0189] In the preferred embodiment, a function, hereinafter called MN_SetFilePath( ), sets the current file path associated with a MitoMine™ parser.

[0190] In the preferred embodiment, a function, hereinafter called MN_GetFilePath( ), gets the current file path associated with a MitoMine™ parser.

[0191] In the preferred embodiment, a function, hereinafter called MN_SetCustomContext( ), may be used to get the custom context value associated with a given MitoMine™ parser. Because MitoMine™ itself uses the parser context (see PS_SetContextID), it provides this alternative API to allow custom context to be associated with a parser.

[0192] In the preferred embodiment, a function, hereinafter called MN_GetCustomContext( ), may be used to get the custom context value associated with a given MitoMine™ parser. Because MitoMine™ itself uses the parser context (see PS_SetContextID), it provides this alternative API to allow custom context to be associated with a parser.

[0193] In the preferred embodiment, a function, hereinafter called MN_GetParseCollection( ), returns the collection object associated with a parser. MN_SetParseCollection( ) allows this value to be altered. By calling MN_SetParseCollection( . . . ,NULL) it is possible to detach a collection from the parser in cases where you wish the collection to survive the parser teardown process.

[0194] In the preferred embodiment, a function, hereinafter called MN_SetParseCollection( ), returns the collection object associated with a parser. MN_SetParseCollection( ) allows this value to be altered. By calling MN_SetParseCollection( . . . ,NULL) it is possible to detach a collection from the parser. This would be useful in cases where it is preferable to permit the collection to survive the parser teardown process.

[0195] In the preferred embodiment, a function, hereinafter called MN_GetMineLanguageTypeDB( ), returns a typeDB handle to the type DB describing the structures utilized by the specified mine language. If the specified typeDB already exists, it is simply returned, otherwise a new type DB is created by loading the type definitions from the designated MitoMine™ type specification file.

[0196] In the preferred embodiment, a function, hereinafter called MN_KillParser( ), disposes of the Parser database created by MN_MakeParser( ). A matching call to MN_KillParser( ) must exist for every call to MN_MakeParser( ). This call would also invoke MN_CleanupRecords( ) for the associated collection.

[0197] In the preferred embodiment, a function, hereinafter called MN_Parse( ), invokes the MitoMine™ parser to process the designated file. The function is passed a parser database created by a call to MN_MakeParser( ). When all calls to MN_Parse( ) are complete, the parser database must be disposed using MN_KillParser( ).

[0198] In the preferred embodiment, a function, hereinafter called MN_RunMitoMine( ), creates the selected MitoMine™ parser on the contents of a string handle. An parameter could also be passed to the NMN_MakeParser( ) call and can thus be used to specify various debugging options.

[0199] In the preferred embodiment, a function, hereinafter called MN_CleanupRecords( ), cleans up all memory associated with the set of data records created by a call to MN_RunMitoMine( ).

[0200] In the preferred embodiment, a function, hereinafter called MN_RegisterMineMuncher( ), can be used to register by name a function to be invoked to post process the set of records created after a successfull MitoMine™ run. The name of the registered Muncher function would preferably match that of the mining language (see MN_Parse for details). A typical mine-muncher function might appear as follows: 3 static ET_CollectionHdl myMuncher( // My Mine Muncher function ET_MineScanRecPtr scanP, // IO:Scanning context record ET_CollectionHdl theRecords, // I:Collection of parsed records char typeDBcode, 1/ I:The typeDB code charPtr parserType, // I:The parser type/language name ET_Offset root, // I:Root element designator charPtr customString // I:Avail pass cstm strig to muncher ) // R:The final collection

[0201] The ‘scanP’ parameter is the same ‘scanP’ passed to the file filter function and can thus be used to communicate between file filters and the muncher or alternatively to clean up any leftovers from the file filters within the ‘muncher’. Custom ‘muncher’ functions can be used to perform a wide variety of complex tasks, indeed the MitoMine™ approach has been used successfully to extract binary (non-textual) information from very complex sources, such as encoded database files, by using this technique.

[0202] In the preferred embodiment, a function, hereinafter called MN_DeRegisterMineMuncher( ), de-registers a previously registered mine muncher function.

[0203] In the preferred embodiment, a function, hereinafter called MN_InvokeMineMuncher( ), invokes the registered ‘muncher’ function for the records output by a run of MitoMine (see MN_RunMitoMine). If no function is registered, the records and all associated memory are simply disposed using MN_CleanupRecords( ).

[0204] In the preferred embodiment, a function, hereinafter called MN_RegisterFileFilter( ), can be used to register by name a file filter function to be invoked to process files during a MitoMine™ run. If no file filter is registered, files are treated as straight text files, otherwise the file must be loaded and pre/post processed by the file filter. A typical file filter function might appear as follows: 4 static EngErr myFileFilter ( // Scan files and mine if appropr HFileInfo *aCatalogRec, // IO:The catalog search record int32Ptr flags, // IO:available for flag use ET_MineScanRecPtr scanP // IO:Scanning context record ) // R:zero for success, else error #

[0205] In the preferred embodiment, a function, hereinafter called MN_ListFileFilters( ),obtains a string list of all know MitoMine™ file filter functions.

[0206] In order to illustrate how MitoMine™ is used to extract information from a given source and map it into its ontological equivalent, we will use the example of the ontological definition of the Country record pulled from the CIA World Fact book. The extract provided in Appendix B is a portion of the first record of data for the country Afganistan taken from the 1998 edition of this CD-ROM. The format of the information in this case appears to be a variant of SGML, but it is clear that this approach applies equally to almost any input format. The lexical analyzer and BNF specification for the parser to extract this source into a sample ontology are also provided in Appendix B. The BNF necessary to extract country information into a sample ontology is one of the most complex scripts thus far encountered in MitoMine™ applications due to the large amount of information that is being extracted from this source and preserved in the ontology. Because this script is so complex, it probably best illustrates a less than ideal data-mining scenario but also demonstrates use of a large number of different built-in mining functions. Some of the results of running the extraction script below can be seen in the Ontology patent relating to auto-generated UI.

[0207] Note that in the BNF provided in Appendix B, a number of distinct ontological items are created, not just a country. The BNF starts out by creating a “Publication” record that identifies the source of the data injested, it also creates a “Government” record, which is descended from Organization. The Government record is associated with the country and forms the top level of the description of the government/organization of that country (of which the military branches created later are a part). In addition, other records could be created and associated with the country, for example the “opt_figure” production is assigning a variety of information to the ‘stringH’ field of the “mapImage” field that describes a persistent reference to the file that contains the map image. When the data produced by this parse is written to persistent storage, this image file is also copied to the image server and through the link created, can be recalled and displayed whenever the country is displayed (as is further demonstrated in the UI examples of the Ontology Patent). In fact, as a result of extracting a single country record, perhaps 50-100 records of different types are created by this script and associated in some way with the country including government personel, international organizations, resources, poulation records, images, cities and ports, neighboring countries, treaties, notes, etc. Thus it is clear that what was flat, un-related information in the source has been converted to richly interconnected, highly computable and usable ontological information after the extraction completes. This same behavior is repeated for all the diverse sources that are mined into any given system the information from all such sources becomes cross-correlated and therefore infinitely more useful that it was in its separate, isolated form. The power of this approach over conventional data mining technologies is clear.

[0208] The foregoing description of the preferred embodiments of the invention has been presented for the purposes of illustration and description. For example, although described with respect to the C* programming language, any programming language that includes the appropriate extentions could be used to implement this invention. Additionally, the claimed system and method should not be limited to the particular API disclosed. The descriptions of the header structures should also not be limited to the embodiments described. While the sample pseudo code provides examples of the code that may be used, the plurality of implementations that could in fact be developed is nearly limitless. For these reasons, this description is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. It is intended that the scope of the invention be limited not by this detailed description, but rather by the claims appended hereto.

Claims

1) A system for the extraction of data from a variety of sources into a single unifying ontology, comprising:

a) an ontology based environment, such environment including an ontology description language (ODL) and a run-time accessible types system;

b) logically connected thereto, an extensible, parsing environment, wherein such parsing environment supports customized reverse-polish plug-in operators;

c) logically connected thereto, a configurable outer parser capable of accepting a BNF (or equivalent) specification describing the source data format;

d) an embedded inner parser capable of executing statements and performing actions directly on the objects and types described by the system ontology.

2) The system of claim 1, further comprising a memory system logically connected thereto, for storing and managing persistent data being processed by the system.

3) The system of claim 1, wherein the ontology-based environment is the ontology environment described in the Ontology Patent.

4) The system of claim 1, wherein the run-time accessible types system is the run-time accessible types system described in the Types Patent.

5) The system of claim 1, wherein the parsing environment is the parsing environment described in the Parse Patent.

6) The system of claim 2, wherein the memory system uses the memory system and model described in the Memory Patent.

7) The system of claim 1, wherein the outer parser is capable of accepting a BNF specification containing specifications for embedded plug-ins.

8) The system of claim 7, wherein the outer parser is capable of accepting a BNF specification containing specification(s) for embedded plug-ins that are passed textual strings consisting of interpretable source statements.

9) The system of claim 7, wherein the embedded inner parser is capable of being invoked by one or more the plug-in(s) in the outer parser, wherein the specification for such plug-ins is stored in the BNF specification.

10) The system of claim 1, further comprising a line processor function, wherein such function permits pre-examination and alteration of the lines appearing in the source data before such data is processed by the parser.

11) The system of claim 2, further comprising one or more post-processor functions (munchers), wherein such munchers are registered on the system and are able to perform arbitrary operations on a collection of extracted ontological records prior to their instantiation into persistent storage by the memory system.

12) The system of claim 1, wherein both parsers are logically connected to a common environment, wherein such comment environment contains a set of assignable and readable registers whose type adapts automatically to any data item assigned to such registers and which registers are persistent in the common enviroment so that they may be used to store the current state of the parsers.

13) The system of claim 1, wherein system can be invoked to mine data from a given source based on actions in a user interface.

14) The system of claim 1, wherein the system can be registered with a running server process connected to a data source in order to allow that server to extract ontological information from the data source on a continuous un-attended basis.

15) A method for the extracting data from a variety of sources into a single unifying ontology, comprising the steps of:

a) receiving source data;

b) parsing the source format with an outer parser, wherein such outer parser includes an embedded parser for an interpreted ontology descriptions language (ODL);

c) parsing the source data with the outer parser and embedded parser using the parsed source format;

d) passing statements in an embedded language to the embedded parser;

e) responsive to one or more actions by the outer parser, executing one or more statements in the embedded language.

16) The method of claim 15, wherein the step of parsing includes the step of receiving a BNF specification.

17) The method of claim 16, wherein the step of parsing includes the step of receiving a BNF specification that includes specifications for embedded plug-ins.

18) The method of claim 17, further comprising the step of passing one or more textual strings to the embedded plug-ins specified in the received BNF specification.

19) The method of claim 18, wherein the step of passing one or more textual strings to the embedded plug-ins includes passing interpretable ODL statements.

20) The method of claim 18, further comprising the step of executing one or more statements ODL statements.

21) The method of claim 20, wherein the step of executing one or more statements comprises ordering the execution of such statements based on the progress of the step of parsing.

22) The method of claim 15, further comprising the step of altering the source data before such data is parsed by the parsers.

23) The method of claim 15, further comprising the step of creating a collection of ontological records.

24) The method of claim 23, further comprising the step of instantiating the collection of ontological records created by the method into persistent storage.

25) The method of claim 24, further comprising the step of registering one or more munchers on the system, wherein such munchers are able to perform arbitrary operations on the collection of ontological records prior to their instantiation into persistent storage

26) The method of claim 15, further comprising the step of registering the method with a server process connected to a data source, such that the server process is able to extract ontological information from the data source on a continuous un-attended basis.