Service Recommendation Based on Targeted Reconstruction of Service Descriptions

Abstract

With the rapidly increasing number of services, there is an urgent demand for service recommendation algorithms that help to automatically create mashups. However, most traditional recommendation algorithms rely on the original service descriptions given by service providers. It is detrimental to the recommendation performance because original service descriptions often lack comprehensiveness and pertinence in describing possible application scenarios, let alone the possible language gap existing between service providers and mashup developers. To solve the above issues, a novel method of Targeted Reconstructing Service Descriptions (TRSD) for a specific mashup query is proposed, resorting to the valuable information hidden in mashup descriptions. TRSD aims at introducing mashup descriptions into service descriptions by analyzing the similarity between existing mashups and the specific query, while leveraging service system structure information. Benefit from this approach, missing application scenarios in original service descriptions, query-specific application scenario information, mashup developers' language habits, and service system structure information are all integrated into the reconstructed service descriptions. Based on the reconstructed service description by TRSD, a new service recommendation strategy is developed. Comprehensive experiments on the real-world data set from ProgrammableWeb.com show that the overall MAP of the proposed TRSD model is 6.5% better than the state-of-the-art methods.