Context-oriented Collaborative Knowledge Retrieval Technology Of Deep Web

With the development of the internet, especially after web2.0, most of these data is in the Deep web. As the explosive growth of the interactive applications such as e-commerce in C hina, C hinese reviews becomes an important part of the Deep web knowledge retrieval, so research on the reviews knowledge retrieval field will become great significance. Traditional information retrieval method which based on the keyword matching is not able to meet the user's specific information and knowledge demand. Therefore, knowledge retrieval emerged and got more and more attentions. However, there is few existing knowledge retrieval technology, and most are based on the huge knowledge base, it is too complicated for the specific areas, and the efficiency is greatly reduced. It needs a new way of knowledge extraction method for knowledge retrieval of reviews.The basic idea of knowledge extraction framework for Chinese reviews is extracting knowledge from their own comments. This study proposes a new concept: K nowledge space, which is the knowledge collection of products in this paper. This framework main contains into two parts: building the knowledge space, using a method based on skip-gram model to train review data and build the knowledge space in this part. The algorithm does not need any feature extraction and modeling for the review data. Thus, it is applicable for massive data, and conveniently extends to parallel processing; retrieving knowledge and optimization results, using Euclidean distance to find the knowledge which is closely linked to the query and use 2-gram algorithm to optimize the results. Based on Hierarchical softmax to constructing knowledge space fast algorithm is the optimization part of the process of construction of knowledge space, which the gradient decent method in the application of mathematical theory, each selection is the best path.The knowledge retrieval framework based on Skip-gram neural network learning model for processing reviews not only has a good performance in reaction time and accuracy, more excellent in user experience. This framework provides to users is more extensive knowledge about the product, which is better adapted to users ' needs than the traditional retrieval method which is based on the keyword matching, provides more information and is a better choice to users.