Snippet Generation Using Local Alignment for Information Retrieval (LAIR)

This paper proposes the method LAIR, which is an adaptation of the local sequence alignment method, commonly used in bioinformatics to identify similarities between gene sequences, for information retrieval in snippet generation. This adaptation draws parallels between detecting similarity in DNA and gene sequences and the query search process. Documents are likened to target gene sequences, and queries to gene query sequences. Once the similarity is established, local sequence alignment is used in snippet generation, similar to its application in bioinformatics. Automated quality measures that focus on informativeness and readability are used to assess the quality of the generated snippets. When comparing the outcomes with Google snippets, LAIR demonstrates competitive performance across evaluation metrics for informativeness and readability. Evaluation metrics include snippet length, number of fragments in the snippet, number of query terms in the snippet, ratio of missed web pages. In addition, the benefits of the local sequence alignment method are emphasized by adapting this method to query searches in YouTube videos as well. This improvement enables user to obtain the relevant section with query words and watch only this part instead of watching the entire video.