Enriching case descriptions using trails in conversational recommenders

Case based recommenders often use similarity as a surrogate for utility. For a given user query, the most similar products are given as recommendations. Similarities are designed in such a way that they closely approximate utilities. In this paper, we propose ways of estimating robust utility estimates based on user trails. In conversational recommenders, as the users interact with the system trails are left behind. We propose ways of leveraging these trails to induce preference models of items which can be used to estimate the relative feature specific utilities of the products. We explain how case descriptions can be enriched based on these utilities. We demonstrate the effectiveness of PageRank style algorithms to induce preference models which can in turn be used in re-ranking the recommendations.