Making Search Conversational to Improve Knowledge Access

Implementing taxonomies to support conversational search reflects the fact that organizations are “seeing there may be value in helping direct the similarity by having a structured set of words, nodes, and, in some cases, it can be sentence vectors. As Osborne pointed out, “It allows us to have an organized structure that helps direct the search, just like we did for decades.” Additionally, taxonomies support multiple varieties of search, including semantic search, keyword- based search, similarity search, dissimilarity search, query-based search, and hybrid representations of these types of search. They formalize the knowledge language models need to make their conversational actions worthwhile. “Taxonomies are a very compact representation of knowledge of different concepts and the relationships in a corpus,” Allen denoted. “They help the LLM find documents that may be more relevant because taxonomies express these implicit relationships between documents and a corpus through concept mapping.”

The Path Forward

Conversational search is extolled for its ability to transform search into an interactive, natural language process in which users can get answers in an iterative fashion, through refining the original search query, rather than sifting through a list of links. However, as numerous subject matter experts have made clear, there are still situations in which organizations need the underlying documents—even if it’s just to verify the results of language models. As such, hybrid paradigms involving both traditional and modern search constructs will likely continue to be employed to produce the most value for applying search to KM.

According to Allen, “Semantic search is interesting and powerful, but I think its prominence as its own thing is going to be short-lived. LLMs aren’t just enabling quicker information retrieval, they’re helping to enable, with agents and other adaptations of LLMs, other deployments, different architectures, and content generation to perform more steps in a process.”

In essence, making search conversational is about crafting a search experience that feels like a natural dialogue with a helpful and knowledgeable assistant, providing relevant and personalized results through intuitive interactions powered by advanced AI, vectorization, semantic search, and machine learning technologies.