Query Refinement in Similarity Retrieval Systems

In many applications, users specify target values for certain attributes/features without requiring exact matches to these values in return. Instead, the result is typically a ranked list of the top k objects that best match the specified feature values. User subjectivity is an important aspect of such queries, i.e., which objects are relevant to the user and which are not depends on the perception of the user. Due to the subjective nature of similarity-based retrieval, the answers returned by the system to a user query often do not satisfy the user’s information need right away; either because the weights and the distance functions associated with the features do not accurately capture the user’s perception or because the specified target values do not fully capture her information need or both. The most commonly used technique to overcome this problem is query refinement. In this technique, the user provides to the system some feedback on the “relevance” of the answers to the user’s query. The system then analyzes the feedback, refines the query (i.e., modifies the weights, distance functions, target values etc.) evaluates it and returns the new results. In this paper, we provide an overview of the techniques used to construct the refined query based on the feedback from the user as well as the techniques to evaluate the refined query efficiently. We present experimental results demonstrating the effectiveness of the techniques discussed in the paper.