Exploration of Opinion-aware Approach to Contextual Suggestion.

摘要

In this paper we describe our effort on TREC Contextual Suggestion Track. Using resources such as description or websites of example suggestions to build user profile has been proven to be effective on last year's TREC. This year we also leverage the power of using user profile. Real world opinions of the suggestions are used in our method to build both user profile and candidate suggestion profile. Two ranking method are investigated to rank the candidate suggestions: linear interpolation and learning to rank. For description generation, we apply the similar method as used in the last year. The structured description combines the category information of the suggestion, meta-description of the website, reviews of the suggestion and the similar example suggestions that the user liked. Official results of our submitted runs show the effectiveness of the proposed method.