Enhancing Object Distinction Utilizing Probabilistic Topic Model

We develop a novel method for enhancing object distinction problem, which is to distinct different objects with identical names. The traditional approach is based on the attributes or contexts of the objects, not using the information of some attribute which may propagate between the objects and the attributes. To this end, we propose an approach which uses probabilistic topic model to improve the accuracy. We first model the attribute which has unstructured information and get its topics which may propagate to object as a new attribute, and then calculate the similarity values of the context attributes of the two objects with identical names, then we use these similarity values to build a decision tree model based on the training set. For the problem of object distinction for people, we combine the affiliation similarity with other context attributes similarity to judge whether the two people who share the same name correspond to the same people in real life. Experiments show that our method using Probabilistic Topic Model can achieve high accuracy.