Research and Application of Automated Search Engine Based on Machine Learning

At present, most search engines mostly use keyword matching mode, without considering the inherent semantics of the text, resulting in low accuracy and recall rates of search results. In order to solve this problem, topic models have been widely used in information retrieval due to their advantages in mining document hidden concepts, but there are still the following problems: First, its unsupervised learning form causes the generated topics to be unclear in interpretability; Second, the topic model is more suitable for processing long texts, and is not suitable for query sentences with fewer features. In order to solve the above problems, this paper proposes an intelligent retrieval method based on topic model based on the semantic characteristics of intelligent retrieval. This method conducts research from two perspectives of document retrieval model and user questioning. It uses topic model and clustering algorithm to mine the latent semantics of documents and constructs a document index. It uses ontology technology to process user question sentences and standardize retrieval styles. Firstly, expert knowledge is selected as the prior knowledge of the topic model, which improves the defect that the topic of the unsupervised topic model deviates from the actual semantics, realizes the semi-supervised topic modeling, and generates the document topic tags that meet the characteristics. Secondly, it studies the clustering method of documents, combines Bagging ensemble learning ideas, and improves the traditional k-means algorithm to solve the problem of sparse text topic distribution generated by topic model training on the original corpus. Third, construct the domain ontology, describe the logical relationship between each entity, and complete the persistence of the ontology, identify the semantic and logical relationship of the query sentence in the retrieval phase, reconstruct the user retrieval conditions, and discover the user's real retrieval requirements; finally, according to The comprehensive matching rate of the query and the document, and the result list is sorted according to the similarity to improve the retrieval accuracy and recall rate.