Topic-Aware Automatic Snippet Generation for Resolving Multiple Meaning on Web Search Result

Abstract

In recent years, the amount of information on the Web is growing exponentially with the spread of the Internet. We generally use search engines to search for the intended information. However, the search engine displays the Web pages including the entered search query in list format. It is difficult for the user to find out the intended information if the entered search query is a word whose meaning depends on the situation and location of the user. It needs the intended information to the multiple hidden topics. In this research, we classify Web search results based on each topic. The topic is defined as the latent meaning, and the contents included in the word. Moreover, our method displays automatically generated snippets for each topic with the Web search results to the user. It is easy to find required information from snippets, even though the intended information is ambiguous. It first classifies the Web search results by Latent Dirichlet Allocation (LDA) which is a major topic model method. It then generates the snippets using Conditional Variational AutoEncoder (Conditional VAE) based on the clustering of We search results. It is expected that using LDA for the clustering will group the Web search result according to the latent meanings of the search query. Also, we expect that proper snippets will be generated for each topic by Conditional VAE. In this paper, we show that LDA is effective for the clustering of Web search results. Moreover, the snippets generated by Conditional VAE is able to generate sentences considered each topic.