A Research About The Pattern Acquisition For Free Text IE

To acquire the event IE patterns automatically is the key to the improvement of information extraction (IE) system's portability. Using a general concept knowledge base such as WordNet/HowNet in the pattern acquisition process will further decrease the labor and skill requirement to the user. This dissertation provides such a pattern automatic acquisition mechanism called GenPAM (General Concept KB-based IE Pattern Acquisition Method), which is mainly composed of a particular pattern representation method, a series of pattern acquisition steps, and a pattern learning algorithm GPLA (IE Pattern Learning Algorithm). GPLA in GenPAM can learn the general patterns from the specific patterns automatically with the support of WordNet/HowNet and can do word sense disambiguation (WSD) implicitly during the pattern learning process. The experiment done using GenPAM indicates that this mechanism can largely decrease the labor and the skill requirement to the user during the pattern acquisition process and the patterns acquired through GenPAM are well enough to guide the factual event IE.Besides acquiring the event IE patterns, acquiring the relational IK pattern is also important and necessary to many applications such as open-domain question answering. This dissertation provides such a relational IE pattern learning method called BRPAM, which can acquire some patterns based on a few relations initially provided by the user, and use these acquired patterns to get more relations, and then use these more relations to get more patterns, and so on, until a terminating condition is satisfied. The terminating condition may be a predefined fixed number of iterations or there are no new patterns or relations produced.BRPAM has been used to acquire the patterns and relations of a bi-relation class from Web pages and the patterns and relations of a bi-relation class from free texts to explain its factual application. The experiment done using...