Preference enhanced hybrid expertise retrieval system in community question answering services

摘要

Here, we propose a preference enhanced hybrid expertise retrieval (PEHER) system in community question answering services. PEHER consists of three segments, namely, preferability estimator, authority estimator, and expertise estimator. The preferability estimator utilizes the textual information to determine both intra-profile and inter-profile preferences of answerers for each term. The intra-profile preferences consider the preference of a term using the answering history of a given answerer. The inter-profile preferences incorporate the preferences of all answerers for a term. These preferences are then used to determine the preferability of each answerer for each of the archived questions. The authority estimator considers the textual familiarity between each archived question and the profile of each answerer as the weight of the associated link in the network. The expertise estimator is composed of three blocks, namely, question similarity finder, proficiency estimator, and expert list generator. The question similarity finder finds the similarities between the new question and each of the archived questions. The proficiency estimator uses the said similarities of the archived questions along with their preferabilities to decide the proficiencies of answerers for the new question. Finally, the expert list generator considers the authorities and proficiencies to generate a list of experts for a given question. We compare PEHER with twenty existing methods on four real-world datasets using five performance measures. We find that PEHER outperforms the comparing algorithms in 92.00% (368 out of 400) cases.