Integration Of Social Relations Attribute Map Clustering Expert Disambiguation Method

Under the background of big data, a lot of the same name experts corresponding documents and pages brought deviations and difficulties on queries in the retrieval process, so the same name experts’ disambiguation was important, which had also become one of the basic and important hotspots in information retrieval research. Rich contents were found in the expert page documents from massive data, in addition to personal attributes of gender, age, occupation, organization, research, etc. there also social associations between an expert and co-workers existed in the experts documents in a particular field or hobby. The attribute information had certain effects in judging the experts’identity. Based on this, we focused on social relations of experts, and achieved the experts disambiguation task by dividing the experts’social network to clusters. Specifically, we completed this paper and achived the goal from the following aspects:We proposed a method based on expert’s attributed graph clustering model. First, we found the co-occurrence between experts and other characters under a limited set of attributes. Second we augmented experts attribute node and built the experts attribute graph based on graph theory to combine the expert attribute consistency and structure consistency together, then establish an entropy model to measure both attribute information and structural information model and calculate super nodes and super edges of their entropy, by minimizing the entropy to get the best clustering and division. The experiment results show that the proposed method based on attribute graph clustering achieved good effect.We combined the expert social relations with attributed graph and proposed an expert’s disambiguation models based on attributed graph clustering which fused with expert’s social relationship. Actually this method is to improve the attributed graph clustering by using the community of experts. We found the direct connection between experts and co-workers in disambiguation document set and used the graph kernel to convert the connection to multiple expert’s communities. The communities were then combined with the attributed graph by which we used to achieve the disambiguation task. Experimental results show that the method with expert social relations performed better than the method without expert social relations and also normal cut graph clustering.The research focused on expert’s attribute relations and social relations which is the base we constructed a framework and did experiments on collected expert’s disambiguation data, finally built a disambiguation prototype expert system.