38results about How to "Realize automatic mining" patented technology

The invention provides a method and a device for excavating a badcase (badcase) of a search engine, wherein the method comprises the following steps of a preprocessing procedure: extracting a certain number of sessions as samples from a session (session) log, and extracting a feature vector describing the search quality from each session of the samples; clustering the samples by utilizing the feature vector of each session; determining confidence coefficient of each category obtained by clustering the samples, wherein the confidence coefficient represents the low degree of the search quality; an excavating procedure: determining an action sequence in the same query in a session log to be excavated, and extracting a feature vector describing the search quality from the action sequence; determining the category of the query by computing the distance between the feature vector of the query and the feature vector of each category; if the confidence coefficient of the category of the query is beyond a preset high threshold, determining that the search engine has the badcase to the query. According to the method and device for excavating the badcase of the search engine, which are disclosed by the invention, the automatic excavation of the badcase of the search engine can be realized, so that the badcase of the search engine can be timely and exactly found out.

The present application discloses a negative sample mining method, device, electronic device and storage medium, and relates to the fields of artificial intelligence such as deep learning and natural language processing. The method may include: when using at least two graphics processors to perform multi-machine model training, Any graphics processor performs the following processing: in any training batch, each query in the training batch is regarded as a pending query; for each pending query, the other queries in the training batch except the pending query are respectively Each query and each query in the training batch corresponding to each graphics processor except itself are used as an associated query, and the sample corresponding to each associated query is used as a weak negative sample corresponding to the query to be processed, and the corresponding samples include positive samples and / or or strong negative samples. By applying the solution described in this application, the model training effect and data utilization rate can be improved.