The Application of Neural Network in Information Retrieval

摘要

Neural network offers a new perspective in Information retrieval systems once it came into being, with its excellent performance of adaptability, learning ability, fault tolerance and parallel processing. This paper introduces the definition, advantages and classification of neural network at first, then analyzes several representative applications of neural network models and related algorithms in information retrieval systems such as the application of Spreading activation networks, COSIMIR model, Kohonen network SOFM and CC4 network. In succession summarized why until now the neural network models are merely used in the practical IR systems, and concluded that some bottlenecks are on the way in the extension of these applications: one reason is that it is hard to select proper exercise sample data and to set weighted importance which are the key steps for an ideal result in neural network. The other is that it is time consuming to handle large vectors with fulltext indexing, large dimension spaces as they commonly occur in IR require either expensive hardware or a considerable amount of time to know. At last introduced three methods to solve the high dimension problem, especially interpreted the Latent Semantic Indexing.