Creative query suggestion
Remember the time when you didn’t have Google to google something? We have come very far, from using search engines as libraries to question answering frameworks. As search engines improve, the complexity of user queries also increases. People often have subtle information needs, such as to make sense of a knowledge space and to find paths through that space. In such cases, you may not be able to formulate a precise query or there may be no definitive answer that would satisfy the your information need. For example, when you start researching a new topic, you may not know the technical terms of the field. In this work, we aim to retrieve creative suggestions that are surprising to the user while retaining relevance to the user’s question. The motivation is to draw your interest to peripheral scenarios and to stimulate your imagination.
We define two metrics Relevance and Diversity to measure how much the suggestion set is related to the search intent of the user and how different the suggestions are from the search intent of the user and other suggestions in the suggestion set. We contribute a new framework based on these metrics to evaluate the quality of the suggestions as well as a content-based approach for producing suggestions. We propose an evaluation methodology based on Microsoft’s Machine Reading COmprehension (MS MARCO) query answering dataset. Results from a user evaluation indicate that our method produces suggestions where the users’ judgments about relevance corresponds to the Relevance metric and users’ judgments about surprise correspond to the Diversity metric.
Here’s an example of the suggestion we identified: Example query: Why should alcohol be banned? We identify an alternative query “what impact did prohibition have on the development of organized crime.” The alternative is not exactly similar to the input scenario but is based on it and benefits from knowledge of it. The transition from banning alcohol to an alternative that links the historical 18th U.S. constitutional amendment (better known as Prohibition) to organized crime guides the search in a new direction.
Writing and defending a master’s thesis was a unique experience. I got a flavour of real research: defining problem statements, writing well-formed arguments, crisply describing methods and results, and conducting a user study. Working under the mentorship of my professor and learning from his experience still helps in the projects I work on.
Did you know? The word “google” is recognized by the Oxford dictionary as a transitive verb. What a product!
