Browse Prior Art Database

Including Key Terms from Watched Hypotheses and Passages in User Profiles for use in Query Formation to Bias Search in Direction of Demonstrated Customer Intent

IP.com Disclosure Number: IPCOM000250419D
Publication Date: 2017-Jul-12
Document File: 3 page(s) / 139K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a system for a search engine that uses a watched hypothesis in a systematic effort to bias a search in favor of material the user has previously taken affirmative steps to flag as relevant.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 45% of the total text.

1

Including Key Terms from Watched Hypotheses and Passages in User Profiles for use in Query Formation to Bias Search in Direction of Demonstrated Customer Intent

Current powerful search engine techniques can, when mixed with enough computing power, both locate salient passages and hypotheses (short answers), and rank the same by relevance for the user. Nonetheless, in an enormous corpus, finding all results remains a challenge. This is especially true if the user asks a question with imprecise language or if the user is unsure exactly how to explain the object of the search. Anything boosting the quality of the information brought to bear on the search is helpful in the process of identifying the best material to return to the user. Recent developments in cognitive QA systems enable the user to take an active role in communicating more information to the search engine. Watched questions is a feature to automatically repeat a saved question following ingestion of new documents to see if anything of user interest has changed. Watched hypotheses allow a user to identify as interesting returned hypotheses and passages; this information has been used by existing systems in a variety of ways. However, the challenge in any QA system is discovery and ranking of passages so that the most relevant material is made available to the user.

The disclosed system leverages the extra information communicated by watched passages and watched hypotheses to build an enhanced customer profile. The search system then uses this profile in augmentation of relevant queries to bias a search in favor of documents related to customer interests. The core novelty is the suggested use of a watched hypothesis in a systematic effort to bias a search in favor of material the user has previously taken affirmative steps to flag as relevant.

This approach extends the ideas of implicit feedback, where the system infers user interest/intent (e.g., user has opened a document or followed a link). This process infers interest from a separate user action; it is called passive because the user intent is referred from an action that is not necessarily driven by the intention imputed by the system. It is a proven strategy in net search and online retailing. Designation of a hypothesis or passage for watching is a far more affirmative act to indicate user interest or user assessment of relevance; as a result, the action taken in response to it can be correspondingly more aggressive.

The novel system uses the watched hypotheses/passages to construct a user profile that can be accessed for query augmentation. By adding key terms from the annotation to the query in a search for topics related to the watched hypothesis, the search can be biased in favor of the topics of demonstrated importance to the user.

2

Following are the process steps: (Steps 1-3 are known; steps 4-8 are novel):

1. A user submits a question to a Deep QA system. 2. The Deep QA system processes the user question and return...