ReLiC: Entity Profiling by using Random Forest and Trustworthiness of a Source - Technical Report
This addresses the need for accurate and current entity profiles from web data, which is incremental as it builds on existing entity resolution techniques.
The paper tackles the problem of entity profiling by merging data from multiple web sources to build complete and up-to-date profiles, handling incorrect or obsolete data. The proposed ReLiC system uses a two-phase method with random forest for entity resolution and trustworthiness of sources, and it outperforms state-of-the-art methods in experimental results.
The digital revolution has brought most of the world on the world wide web. The data available on WWW has increased many folds in the past decade. Social networks, online clubs and organisations have come into existence. Information is extracted from these venues about a real world entity like a person, organisation, event, etc. However, this information may change over time, and there is a need for the sources to be up-to-date. Therefore, it is desirable to have a model to extract relevant data items from different sources and merge them to build a complete profile of an entity (entity profiling). Further, this model should be able to handle incorrect or obsolete data items. In this paper, we propose a novel method for completing a profile. We have developed a two phase method-1) The first phase (resolution phase) links records to the queries. We have proposed and observed that the use of random forest for entity resolution increases the performance of the system as this has resulted in more records getting linked to the correct entity. Also, we used trustworthiness of a source as a feature to the random forest. 2) The second phase selects the appropriate values from records to complete a profile based on our proposed selection criteria. We have used various metrics for measuring the performance of the resolution phase as well as for the overall ReLiC framework. It is established through our results that the use of biased sources has significantly improved the performance of the ReLiC framework. Experimental results show that our proposed system, ReLiC outperforms the state-of-the-art.