Search

With the rapid development of Internet, how to extract personal relations from Internet has become an important research topic in information extraction. However, current relation extraction researches mainly focus on the processing of English language, the researches focus on Chinese are less. At the same time, there are two main problems in current personal relation extraction approaches: 1) it is difficult to get a large amount of high quality training data without manually label effort; 2) the performance of personal relation extraction from Chinese is unsatisfactory. To solve the first problem, we propose an improved distant supervision method which is applied to Chinese language and can label large-scale of high quality training data automatically. To solve the second problem, we extract three features based on word embedding and combine them with the basic features. In the experiment, the improved distant supervision methods improve the quality of training data significantly. And the word embedding features improve the average F1 score by 4% than the basic features.

The key of big text documents data analysis is to classify those text documents. To classify those text documents, it is necessary to represent those text documents as vectors which is vector space model (VSM). A powerful vector space model should remain the classification information with dimensions as little as possible. To achieve that, it is important to select most effective features for text classification. Unlike the supervised selection method which utilizes the category information in the training data, we propose an unsupervised feature selection method. Our method requires no category information which makes our method has more application scenarios as the labeled data is expensive and inaccurate. Unlike other unsupervised methods, our method utilizes word embedding to find the words with similar semantic meaning. The word embedding maps the words into vectors and remains the semantic relationships between words. We select the most representative word on behalf of the words with similar semantic meaning because it is redundant to include all those words as features. We demonstrate on Reuters-21578 dataset, that our method outperforms other methods. Especially, our method has great advantage when select limited features.

The text representation is fundamental for text mining and information retrieval. The Bag Of Words (BOW) and its variants (e.g. TF-IDF) are very basic text representation methods. Although the BOW and TF-IDF are simple and perform well in tasks like classification and clustering, its representation efficiency is extremely low. Besides, word level semantic similarity is not captured which results failing to capture text level similarity in many situations. In this paper, we propose a straightforward Bag Of Word cLusters (BOWL) representation for texts in a higher level, much lower dimensional space. We exploit the word embeddings to group semantically close words and consider them as a whole. The word embeddings are trained on a large corpus and incorporate extensive knowledge. We demonstrate on three benchmark datasets and two tasks, that BOWL representation shows significant advantages in terms of representation accuracy and efficiency.