Abstract: The system and method of the present invention automatically assigns “scores” to the predictor/variable value pairs of a conventional probabilistic model to measure the relative impact or influence of particular elements of a set of topics, items, products, etc. in making specific predictions using the probabilistic model. In particular, these scores measure the relative impact, either positive or negative, that the value of each individual predictor variable has on the posterior distribution of the target topic, item, product, etc., for which a probability is being determined. These scores are useful for understanding why each prediction is made, and how much impact each predictor has on the prediction. Consequently, such scores are useful for explaining why a particular prediction or recommendation was made.

Abstract: A system to facilitate trading of advertising comprises a publisher broker representing at least one publisher and to determine an ask for an advertisement space on the publisher's webpage, an advertiser broker representing at least one advertiser and to manage an advertiser's bid for the advertisement space, and an exchange to facilitate a transaction for the advertisement space between the publisher broker and the advertiser broker. A method of facilitating trading of advertising comprises receiving an ask from a publisher broker for advertisement space on a webpage, receiving a bid from an advertiser broker for the advertisement space, and pairing the ask with the bid. A method for enriching user information comprises aggregating user information about a user, storing the aggregate user information according to a user identifier, receiving the user identifier from an exchange, and sending the aggregate user information to the exchange.

Abstract: A publisher union comprises a plurality of publishers, a channel, and a publisher union administrator. The publisher union is administered by receiving a channel proposal, determining whether the channel proposal is acceptable, forming a channel, and presenting the channel for monetization. User information is gathered by the publisher union by establishing a domain, collecting user information, aggregating the user information, and providing the aggregated user information to publisher union members.

Abstract: Architecture for integrating and generating back-off grammars (BOG) in a speech recognition application for recognizing out-of-grammar (OOG) utterances and updating the context-free grammars (CFG) with the results. A parsing component identifies keywords and/or slots from user utterances and a grammar generation component adds filler tags before and/or after the keywords and slots to create new grammar rules. The BOG can be generated from these new grammar rules and can be used to process the OOG user utterances. By processing the OOG user utterances through the BOG, the architecture can recognize and perform the intended task on behalf of the user.

Abstract: A system and method for prediction of a user goal for command/control of a personal device (e.g., mobile phone) is provided. The system employs statistical model(s) that can predict a command based, at least in part, on past user behavior (e.g., probability distribution over a set of predicates, and, optionally arguments). Further, the system can be employed with a speech recognition component to facilitate language modeling for predicting the user goal. The system can include predictive user models (e.g., predicate model and argument model) that receive a user input (e.g., utterance) and employ statistical modeling to determine the likely command without regard to the actual content of the input (e.g., utterance). The system employs features for predicting the next user goal which can be stored in a user data store. Features can capture personal idiosyncrasies or systematic patterns of usage (e.g., device-related, time-related, predicate-related, contact-specific and/or periodic features).

Abstract: Architecture for integrating and generating back-off grammars (BOG) in a speech recognition application for recognizing out-of-grammar (OOG) utterances and updating the context-free grammars (CFG) with the results. A parsing component identifies keywords and/or slots from user utterances and a grammar generation component adds filler tags before and/or after the keywords and slots to create new grammar rules. The BOG can be generated from these new grammar rules and can be used to process the OOG user utterances. By processing the OOG user utterances through the BOG, the architecture can recognize and perform the intended task on behalf of the user.

Abstract: An architecture is presented that leverages discrepancies between user model predictions and speech recognition results by identifying discrepancies between the predictive data and the speech recognition data and repairing the data based in part on the discrepancy. User model predictions predict what goal or action speech application users are likely to pursue based in part on past user behavior. Speech recognition results indicate what goal speech application users are likely to have spoken based in part on words spoken under specific constraints. Discrepancies between the predictive data and the speech recognition data are identified and a dialog repair is engaged for repairing these discrepancies. By engaging in repairs when there is a discrepancy between the predictive results and the speech recognition results, and utilizing feedback obtained via interaction with a user, the architecture can learn about the reliability of both user model predictions and speech recognition results for future processing.

Abstract: A generic predictive argument model that can be applied to a set of slot values to predict a target slot value is provided. The generic predictive argument model can predict whether or not a particular value or item is the intended target of the user command given various features. A prediction for each of the slot values can then be normalized to infer a distribution over all values or items. For any set of slot values (e.g., contacts), a number of binary variables are created that indicate whether or not each specific slot value was the intended target. For each slot value, a set of input features can be employed to predict the corresponding binary variable. These input features are generic properties of the contact that are “instantiated” based on properties of the contact (e.g., contact-specific features). These contact-specific features can be stored in a user data store.

Abstract: A predictive model analysis system comprises a receiver component that receives predictive samples created by way of forward sampling. An analysis component analyzes a plurality of the received predictive samples and automatically determines whether a predictive model is reliable at a time range associated with the plurality of predictive sample, wherein the determination is made based at least in part upon an estimated norm associated with a forward sampling operator.

Abstract: The subject disclosure pertains to systems and methods that optimize advertisement campaigns. In particular, total utility that can be derived by an advertiser given particular keywords is maximized. The price of each keyword/slot pair can be determined or estimated and bids adjusted automatically to maximize advertiser utility or return on investment for a campaign.

Abstract: An advertisement display system comprises an analyzer component that analyzes one or more of data associated with at least a portion of a multimedia item, demographic information associated with a user, and contextual data. A presentation component selectively provides at least one advertisement from a plurality of advertisements to a reviewer of the multimedia item based at least in part upon the analysis. The system, for example, can further comprise an ad server, wherein the presentation component receives the at least one advertisement from the ad server based at least in part upon the analysis.

Abstract: Data slices of historical time series are leveraged to facilitate in more accurately predicting like data slices of future time series. Different predictive models are employed to detect outliers in different data slices to enhance the accuracy of the predictions. The data slices can be temporal and/or non-temporal attributes of a data set represented by the historical time series. In this manner, for example, a historical time series for a network location can be sliced temporally into one hour time periods as a function of a day, a week, a month, a year, etc. Outliers detected in these data slices can then be mitigated utilizing the predictive time series model by replacing the outlier with the expected value. The mitigated historical time series can then be employed in a predictive model to predict future web traffic for the network location (and advertising revenue values) with a substantial increase in accuracy.

Abstract: Random samples without replacement are extracted from a distributed set of items by leveraging techniques for aggregating sampled subsets of the distributed set. This provides a uniform random sample without replacement representative of the distributed set, allowing statistical information to be gleaned from extremely large sets of distributed information. Subset random samples without replacement are extracted from independent subsets of the distributed set of items. The subset random samples are then aggregated to provide a uniform random sample without replacement of a fixed size that is representative of a distributed set of items of unknown size. In one instance, a multivariate hyper-geometric distribution is sampled by breaking up the multivariate hyper-geometric distribution into a set of univariate hyper-geometric distributions.

Abstract: Decision trees populated with classifier models are leveraged to provide enhanced spam detection utilizing separate email classifiers for each feature of an email. This provides a higher probability of spam detection through tailoring of each classifier model to facilitate in more accurately determining spam on a feature-by-feature basis. Classifiers can be constructed based on linear models such as, for example, logistic-regression models and/or support vector machines (SVM) and the like. The classifiers can also be constructed based on decision trees. “Compound features” based on internal and/or external nodes of a decision tree can be utilized to provide linear classifier models as well. Smoothing of the spam detection results can be achieved by utilizing classifier models from other nodes within the decision tree if training data is sparse. This forms a base model for branches of a decision tree that may not have received substantial training data.

Abstract: A computer-implemented system that facilitates establishing price(s) associated with items comprises a clustering component that clusters a collection of non-identical items into one or more sets of non-identical items. A pricing component receives one or more of active bids and stored bids and simulates an auction of at least one non-identical item within one of the one or more sets of non-identical items as if the non-identical items within the set were identical to determine price(s) associated with the at least one non-identical item. For example, the items can be at least portions of search terms received by a search engine.

Abstract: A computer-implemented method is provided for controlling placement of ad impressions, corresponding to ads, displayed on a web page. The method includes recording features corresponding to ad impressions. Recording features can include collecting sufficient statistics for a Naïve Bayes model in some embodiments. A statistical algorithm is then used to automatically control placement of ad impressions.

Abstract: The transmission of information during ad click-through is disclosed. In one embodiment, a computer-implemented method selects an ad to be displayed on a web page, as one of a plurality of ads within a current cluster in which each of the ad has a probability to be selected. The method displays the ad on the web page, and then detects activation—for example, click-through—of the displayed ad. The method transmits information to an entity associated with the ad, such as an advertiser, upon detecting click-through or other activation of the ad. In one embodiment, the information transmitted includes information regarding the current cluster.

Abstract: A system and method for online reinforcement learning is provided. In particular, a method for performing the explore-vs.-exploit tradeoff is provided. Although the method is heuristic, it can be applied in a principled manner while simultaneously learning the parameters and/or structure of the model (e.g., Bayesian network model). The system includes a model which receives an input (e.g., from a user) and provides a probability distribution associated with uncertainty regarding parameters of the model to a decision engine. The decision engine can determine whether to exploit the information known to it or to explore to obtain additional information based, at least in part, upon the explore-vs.-exploit tradeoff (e.g., Thompson strategy). A reinforcement learning component can obtain additional information (e.g., feedback from a user) and update parameter(s) and/or the structure of the model.

Abstract: A dialog system training environment and method using text-to-speech (TTS) are provided. The only knowledge a designer requires is a simple specification of when the dialog system has failed or succeeded, and for any state of the dialog, a list of the possible actions the system can take. The training environment simulates a user using TTS varied at adjustable levels, a dialog action model of a dialog system responds to the produced utterance by trying out all possible actions until it has failed or succeeded. From the data accumulated in the training environment it is possible for the dialog action model to learn which states to go to when it observes the appropriate speech and dialog features so as to increase the likelihood of success. The data can also be used to improve the speech model.

Abstract: A simulation environment for adapting a speech model (e.g., baseline model) to a user is provided. The user can interact with a base parametric speech model (e.g., statistical model with learnable parameters such as a Bayesian network) and give positive and/or negative feedback when the dialog system has performed what the user considers to be appropriate and/or inappropriate action(s). From the user feedback, the dialog system learns to take actions customized for the particular user. Speaker-dependent adaptation can be extended to the dialog level by performing maximum likelihood linear regression (MLLR) adaptation simultaneously with dialog personalization. Users are immediately able to observe how their feedback has caused the dialog system to adapt, and can quit training whenever they feel that the dialog system has adapted enough for current purposes.