Linguistically aware link analysis method and system
Example, systems, methods, computer media, and other embodiments for determining relevance rankings for pages identified in a search query is provided. In one example, a computer program product can be configured to identify a candidate set of pages in response to a search query. A content-based relevance rank can be determined for at least one page of the candidate set of pages based on a content of the at least one page. The content-based relevance rank can be adjusted for one or more selected pages from the candidate set of pages by distributing a relevance rank from one or more pages that point to the one or more selected pages.
Latest Oracle Patents:
- System and method to use queue pair 1 for receiving multicast based announcements in multiple partitions in a high performance computing environment
- System and method for providing a hybrid set-based extract, load, and transformation of data
- System and method for determination of recommendations and alerts in an analytics environment
- Voice communication analysis system
- Systems and methods for memory tracing in asset managing systems
This application is a continuation of and claims the benefit from U.S. patent application entitled “Linguistically Aware Link Analysis Method and System”, Ser. No. 09/928,962 filed Aug. 13, 2001, inventor Shamim Alpha, attorney docket number 27252.4 (OID-2000-152-01), which is also assigned to the present assignee.
BACKGROUNDThe Internet, including the World Wide Web (the “Web”) allows access to enormous amounts of information which grows in number daily. This growth, combined with the highly decentralized nature of the Web, creates a substantial difficulty in locating selected information content. Prior art Web search services generally perform an incremental scan of the Web to generate various, often substantial indexes that can be later searched in response to a user's query. The generated indexes are essentially databases of document identification information. Search engines uses these indexes to provide generalized content based searching but a difficulty occurs in trying to evaluate the relative merit or relevance of identified candidate documents. A search for specific content in documents or web pages in response to a few key words will almost always identify candidate documents whose individual relevance is highly variable. Thus, a user's time can be inefficiently spent on viewing numerous candidate documents that are not relevant to what they are looking for.
Some prior search engines attempt to improve relevancy scores of candidate documents by analyzing the frequency of occurrence of the query terms on a per document basis. Other weighing heuristics, such as the number of times that any of the query terms occur within a document and/or their proximity to each other, have also been used. These relevance ranking systems typically presume that increasing occurrences of specific query terms within a document means that the document is more likely relevant and responsive to the query. However, this assumption is not always accurate.
Another method to determine the relevancy of a document is by using link analysis. Generally, link analysis assumes a that if important web pages point to a document, then the document is also probably important or relevant. However, typical link analysis models a user's search for information on the Web as fluid moving between different containers where the webpages are represented by containers and links out of a webpage are represented by connecting conduits with the same diameter. What this model assumes is that users coming to a webpage must leave the webpage by following one of the links from the webpage and users are equally likely to follow any of the links from the webpage. If a page does not refer to any webpage, it is assumed to refer to all the webpages. By solving a steady state solution of the system, the model finds the relative likelihood of finding the user on a webpage if a snapshot of the system was taken. The basic problem with the model is that people are not like fluids.
Rather, people can evaluate the relevance of a webpage for a query. That has two implications on the behavior of the user in the system: 1) users will be likely to stop searching based on the relevance of a webpage, and 2) choosing between two links, users will be more likely to follow a link to the more relevant page.
Based on these implications, there is a need for a relevance ranking system where the probability of not leaving a webpage is a function of the relevance of the webpage, and the probability of following an outgoing link from a webpage is a function of the relevance of all referred webpages and the relevance of the webpage.
BRIEF DESCRIPTION OF THE DRAWINGSIn the accompanying drawings which are incorporated in and constitute a part of the specification, embodiments of the invention are illustrated, which, together with a general description of the invention given above, and the detailed description given below, serve to example the principles of this invention.
One or more embodiments described herein relate to network information retrieval and relevance ranking. In one example, methods and systems can be configured to combine link analysis from web pages and linguistic characteristics of the web pages to obtain relevance rankings for search query results. Relevance rankings can be improved to provide more relevant page information to a user in response to a search query.
In one example, a computer-implemented process/product can be configured to assume that the probability that a user will follow a selected out-going link is not equal between all out-going links from a given page. Rather, some are more likely to be followed than others if the user believes the destination page is relevant to their query. Even if the website does not provide any clue (the text associated with link or url itself) to the visitors about which links are more likely to be relevant, users are still more likely to follow a link that points to a more relevant webpage. If upon following a bad (with inferior content) link, visitors will immediately bounce back to the referrer page and follow another link. Users will be effectively spending more time on a page with better content. That will likely mean that we will find the user on a more relevant referred page even in the absence of a visible clue on the referrer page. Thus, the relevance ranking of one or more examples herein combines link analysis rankings with content relevance rankings to obtain page rankings.
In some example embodiments as described herein, since they combine link analysis rankings with content relevance rankings, the relevance rank of a page will increase based on the number of relevant pages that point to it. In other words, if many highly relevant pages point to a selected page, then the selected page must also be highly relevant.
The following includes definitions of exemplary terms used throughout the disclosure. Both singular and plural forms of all terms fall within each meaning:
“Page”, as used herein, includes but is not limited to one or more web pages, an electronic document, network addresses or links, database addresses or records, or other objects that are identifiable using a search query. “Page” and “document” are used interchangeably.
“Software”, as used herein, includes but is not limited to one or more computer executable instructions, routines, algorithms, modules or programs including separate applications or from dynamically linked libraries for performing functions as described herein. Software may also be implemented in various forms such as a servlet, applet, stand-alone, plug-in or other type of application as known to those skilled in the art.
“Logic”, as used herein, includes but is not limited to hardware, software and/or combinations of both to perform a function.
“Network”, as used herein, includes but is not limited to the internet, intranets, Wide Area Networks (WANs), Local Area Networks (LANs), and transducer links such as those using Modulator-Demodulators (modems).
“Internet”, as used herein, includes a wide area data communications network, typically accessible by any user having appropriate software. This includes the World Wide Web. “Intranet” includes a data communications network similar to an internet but typically having access restricted to a specific group of individuals, organizations, or computers.
Illustrated in
With further reference to
With further reference to
The relevance ranking system 115 is embodied as software and includes software components as described below. The relevance ranking system may be a component within the information retrieval system 110 or may be called and executed externally. Once a candidate set of pages is retrieved, a link structure logic 120 determines the link structure of the pages including the out-going links from each page which become in-coming links to another page. This may be performed by using a spider or web crawler as is known in the art and may be performed dynamically for each candidate set of pages or may be obtained from predetermined link structure information.
With reference to
With reference again to
Once the content relevance values are obtained for each page, a probability logic 130 determines a probability that a user will stay on a given page as a function of the content relevance values. For example, if the content relevance values are between 0 and 100, these values can be directly translated into a corresponding percentage value to give the probability of staying on a given page. For example, if the content relevance value for page C is 30, then the probability of a user staying on page C is set to 30% (0.3). Of course, many different transformations can be used including non-linear relationships between the relevance values for the page and linked pages and the probability of staying on a page.
With reference again to
With further reference to
Determining the link rankings for the remaining links is as follows:
As shown in Equations (3)-(5), the probability of following out-going link C-B equals 0.7*0.5/1.2 which is approximately 0.3. Doing a similar analysis for the remaining out-going links, the probability of following link A-B is approximately 0.3, and following link B-C is 0.5. Thus, the probability of following an outgoing link from a parent page is a function of the relevance of all referred child pages and the relevance of the parent page. It will be appreciated that there are many ways to distribute probabilities based on probabilities of parent and child pages. Other distributions can reflect the page relevance of a parent.
Once an initial determination of page relevance values and out-going link values are determined, a relevance rank adjuster 140 adjusts the content relevance values for each page based on the probability values of the link analysis. For example, the relevance rank for page A is modified based on the relevance rank of pages that refer to page A as a function of the probability of going to page A from any of those pages. In other words, if more relevant pages point to page A, then page A is probably more relevant. Thus, there should be a greater probability that a user will be on page A at any given time in relation to the other candidate pages. Using
PA(being)=PA(staying)*PA(being)+P(link C-A)*PC(being) (6)
which becomes
PA(being)=0.7PA(being)+0.4PC(being)=20/56
and for the other candidate pages:
PB(being)=0.5PB(being)+0.3PA(being)+0.3PC(being)=21/56 (7)
PC(being)=0.3PC(being)+0.5PB(being)=15/56 (8)
where
PA(being)+PB(being)+PC(being)=1 (9)
The set of four equations have three unknowns that are solved using known linear algebra techniques. As shown in Equations (6-9), the probability of being on a page is based on the relevance of the page weighted by the probability of being on that page and a sum of the values from all in-coming links weighted by the probability of being on the parent page. The probability of a user being on a page “Prob(being)” is a probability distribution to all candidate pages, thus, the sum of probabilities is one (1). The “Prob(being)” is an absolute probability whereas the probability of staying on a page is conditional since it is assumed that a user must be on that page.
Of course, there are other ways to use content-based relevance values to vary or adjust the probability of being on or leaving a page other than by the given examples. The fundamental approach includes determining the relevance of a page based on a combination of its content-based relevance value and the relevance of links that point to the page. Thus, if more relevant pages point to a page, its relevance value will be increased.
Illustrated in
With reference to
In response to the search query, the information retrieval system 110 identifies a candidate set of pages from the network that potentially match what the user is looking for. For example, the keywords of the query are matched against a pregenerated database of indexes that point to web pages containing or relating to the keyword. The candidate pages are then received by the relevance ranking system 115 for assignment of relevance rankings (block 310). A content-based analysis is executed for each page to determine a relevance value in view of the search query (block 315). As mentioned previously, the relevance value can be any value that reasonably reflects the relevance of the content or subject matter of a page in relation to the key words of the search query. There are many software programs known in the art that can be used to obtain an initial content-based relevance value for a page.
Once an initial relevance value is assigned for each page, the relevance values are translated to a probability that a user will stay on a given page (block 320). If, for example, the initial relevance values are between 0-100 where 100 means the page is very relevant, a simple translation includes directly relating the relevance value of a page to a probability of staying on the page (e.g. relevance value 70 is translated to a 70% probability of staying). Depending on the type of relevance values used, they may directly corresponded to a percentage value as in the above example, or they may be transformed to fit into percentage values based on a desired formula if there is no one-to-one correspondence. The probability of staying on a page depends on a content-based relevance ranking and topology of the pages (link structure).
With further reference to
Using the example candidate pages from
With reference again to
With the present teachings, relevance rankings can be based on linguistically aware link analysis where link values incorporate content-based relevance values of associated pages as a function of the page link structure. Link analysis rankings can become linguistically aware since they can be combined with content-based relevance values. In one example as described previously, a probability of not leaving a webpage and the probability of following an outgoing link from a webpage are functions of the relevance of referred webpages and the relevance of the webpage. In this manner, improved relevance rankings for web pages can be obtained based on a given search query.
While various examples have been illustrated by the description of embodiments thereof, and while the embodiments have been described in considerable detail, it is not the intention of the applicants to restrict or in any way limit the scope of the appended claims to such detail. Additional advantages and modifications will readily appear to those skilled in the art. For example, the relevance rank system may be a function within the information retrieval system or an external program. The link structure logic may perform the structure analysis dynamically or it may simply obtain link structure information from an external application or source which is available. The same applies to the content analyzer logic. Therefore, the example systems and methods, in their broader aspects, are not limited to the specific details, the representative apparatus, and illustrative examples shown and described. Accordingly, departures may be made from such details without departing from the spirit or scope of the applicant's general inventive concept.
Claims
1. A computer program product for determining a relevance rank for a plurality of pages identified by a search query, computer program product, when executed, causes one or more computers to perform a method, the method comprising:
- identifying a candidate set of pages in response to the search query;
- determining a content-based relevance rank for at least one page of the candidate set of pages based on a content of the at least one page; and
- adjusting the content-based relevance rank for one or more selected pages from the candidate set of pages by distributing a relevance rank from one or more pages that point to the one or more selected pages.
2. The computer program product of claim 1 where the relevance rank from a page that points to the selected page is based on, at least in part, a content-based relevance rank for the page.
3. The computer program product of claim 1 further including obtaining a link structure of the candidate set of pages to determine in-coming and out-going page links.
4. The computer program product of claim 1 where the adjusting includes, for a first page, combining the content-based relevance rank of the first page with the relevance ranks distributed from the one or more pages that point to the first page.
5. The computer program product of claim 1 where the computer program product is embodied in an information retrieval system.
6. A method of determining a relevance rank for a selected page in response to a search query, the method comprising:
- identifying a candidate set of pages in response to the search query;
- determining a content-based relevance rank for at least one page of the candidate set of pages based on a content of the at least one page; and
- adjusting the content-based relevance rank for a selected page from the at least one page of the candidate set of pages by using at least part of a relevance rank from at least one of one or more pages that point to the selected page.
7. The method of claim 6 where the relevance rank includes a content-based relevance rank.
8. The method of claim 6 where the at least part of the relevance rank from the at least one of one or more pages that point to the selected page is distributed to out-going links.
9. The method of claim 6 where the distributing is based on a link structure of the pages including link rank values from in-coming links where the link rank values are determined from distributed values of at least part of the content-based relevance from the one or more pages that point to the selected page.
10. A system for determining a relevance ranking for pages obtained from a network search query, the system comprising:
- link structure logic for obtaining a link structure of the pages which identifies out-going links from each of the pages which become in-coming links to other pages;
- a content analyzer for determining a content of each page;
- a content relevance ranking logic for determining a content relevance rank for each page based on a content of the page in relation to the network query;
- link analysis logic for determining a link ranking for each of the out-going links for each of the pages, where the link ranking of an out-going link from a selected page being based, at least in part, on a content-based relevance rank of the selected page; and
- a relevance rank adjuster for determining and adjusting a relevance rank of a page by combining the content relevance rank of the page with the link rankings associated to in-coming links for the page.
11. A method of ranking a set of candidate pages in response to a search query, the method comprising:
- identifying the candidate pages from a network that potentially match the search query;
- assigning a content-based relevance rank to a selected page from the one or more candidate pages;
- adjusting the content-based relevance rank of the selected page where the content-based relevance rank for the selected page is influenced by a relevance of one or more candidate pages that point to the selected page; and
- ranking the candidate pages based on the adjusted content-based relevance rank.
12. The method of claim 11 where the adjusting is influenced by the relevance including at least part of a content-based relevance rank.
Type: Application
Filed: Dec 22, 2005
Publication Date: May 18, 2006
Applicant: ORACLE INTERNATIONAL CORPORATION (Redwood Shores, CA)
Inventor: Shamim Alpha (Foster City, CA)
Application Number: 11/315,053
International Classification: G06F 17/30 (20060101);