A Text Mining based Method for Policy Recommendation

Abstract

Governmental administrations frequently release various types of policies to enforce specific rules, laws, or economic stimulus plans. For a specific policy, there are many potential target individuals, companies and organizations. However, it is not always timely for a company to get to know the policies that are suitable to it. It is necessary to develop efficient policy recommendation methods to help companies catch up with useful policies right the first time. In this paper, we propose a policy recommendation method based on text mining. This method consists of three phases: 1) Policy structure division; 2) Attribute extraction; 3) Matching and recommendation. Since most of policy texts contain too many contents that might be suitable for multiple different types of companies, policy text is firstly divided into text fragments, and each text fragment is then divided into multiple elementary discourse units (EDUs) which are used as the basic recommendation unit with optimal granularity. Secondly, by combining a Named Entity Recognition (NER) based extractor with a rule-based extractor and a grammar-based extractor, we extract attribute entities and logical relations between these entities from each EDU. Thirdly, on the basis of the extracted attribute entities and logical relations, we calculate the matching score between each fragmented policy text and each company, and these scores are taken as the sorted criteria of policy recommendation. The policy recommendation results cover policy text fragments, related governmental administrations, attribute entities and logical relations, links to the entire policy text, and so on. On the basis of a data set collected from real world, the effectiveness of the proposed method is validated by the experiments of two different scenarios: 1) there is a list of policy texts, and it is needed to recommended suitable policies to a specific company; 2) there is a list of companies, and it is needed to help them to find suitable policy texts. This work can be considered as an intelligent government affairs service for accurately pushing useful policies to target users, and is of great significance to the modern governance system.