Query Generation as Result Aggregation for Knowledge Representation

Knowledge representations have greatly enhanced the fundamental human problem of information search, profoundly changing representations of queries and database information for various retrieval tasks. Despite new technologies, little thought has been given in the field of query recommendation – recommending keyword queries to end users – to a holistic approach that recommends constructed queries from relevant snippets of information; pre-existing queries are used instead. Can we instead determine relevant information a user should see and aggregate it into a query? We construct a general framework leveraging various retrieval architectures to aggregate relevant information into a natural language query for recommendation. We test this framework in text retrieval, aggregating text snippets and comparing output queries to user generated queries. We show that an algorithm can generate queries more closely resembling the original and give effective retrieval results. Our simple approach shows promise for also leveraging knowledge structures to generate effective query recommendations.