722 results about "Recall rate" patented technology

The invention discloses a CT image pulmonary nodule detection system based on 3D full-connection convolution neural network. The detection system comprises the following five steps: constructing a training set data; performing 3D convolution neural network classification network training; performing 3D convolution neural network segmentation network training; carrying out false-positive suppression in which the trained segmentation network and the false-positive are utilized to inhibit the network; and detecting the pulmonary nodule. The technical schemes of the invention can realize the full automatic detection without any human intervention. At the same time, the recall rate of pulmonary nodule detection can be increased effectively; the false-positive focus of infection is reduced considerably; and a pixel level positioning quantitative and qualitative result for the focus-of-infection area of the pulmonary nodule can be obtained.

A method to extract new keywords based on word frequency and multiple grammars is provided, which belongs to the technology field of a natural language processing, and is characterized by extracting the potential models of part of speech of the multiple grammars of the keywords by researching characteristic part of speech of the keywords and adopting computer to assist excavation and taking the models as the basis of the keywords to extract arithmetic. When extracting the new keywords, firstly excavating the multiple phrases in text in accordance with the potential models of part of speech and extract candidate word set of the keywords, and then excavating potential keywords not loading from titles and add the potential keywords to the candidate keyword set. The application brings forward an improved single text word frequency/inverse text frequency value (tf/idf) format, introduces target-oriented characteristics, grades the candidate keywords, obtains the order of the candidate keywords and gives the keywords of news document after optimizing the results. Compared with the traditional keyword extraction method based on single text word frequency/inverse text frequency value (tf/idf), the method has higher recall rate under the condition of the same precision.

A phishing webpage detection method based on Hungary matching algorithm is characterized by firstly extracting the text feature signatures, image feature signatures and general webpage feature signatures of the rendered webpages and more comprehensively depicting the features after access to webpages; and then computing the optimal matching of bipartite graphs by Hungary algorithm to search for the matched feature pairs among different webpage signatures and more objectively measuring the similarity among the webpages on the basis, thereby improving the phishing webpage detection efficiency. The method is also characterized by determining the inside weights of the text features, image features and global image features by utilizing the area under curve and determining the relative weightsamong the text similarity, image similarity and global image similarity during webpage similarity computation by utilizing logarithmic regression analysis. The precision and the recall rate are greatly improved in the method provided by the invention.

The invention discloses a method for identifying opinion leaders in a social network based on topic link behaviors of users, comprising constructing a social network directed graph, amending edge-weight of the network, for each network node i, calculating the number of shortest paths passing through the node i, calculating the length of the shortest paths, weighting edges of the node, and calculating weighted sum B(i). Opinion leaders based on topic tracking are composed of nodes which have a larger value of B(i). The method of the invention adopts a calculation method of link relations in social network user topics, which reduces scale of network nodes and accordingly decreases system overhead and improves algorithm accuracy and recall rate.

The invention discloses an automatic microblog text abstracting method based on unsupervised key binary word extraction. The automatic microblog text abstracting method comprises the steps of preprocessing a microblog; standardizing a binary word; extracting a key binary word based on a mixed TF-IDF (term frequency-inverse document frequency), TexRank and an LDA (local data area); sequencing sentences based on the intersection similarity and a mutual information strategy; extracting abstract sentences based on a similarity threshold value; generating abstract by reasonably combining the abstract sentences. According to the automatic microblog text abstracting method, the binary word is used as a minimum vocabulary unit, and the binary word has richer text information than words, so that the sentences based on the key binary word is higher in noise immunity and accuracy than the sentences based on key word extraction; meanwhile, when the abstract sentences are extracted, the similarity threshold value is introduced to control redundancy, so that the abstract is higher in recall rate. The abstract generated by the method is accurate, simple and comprehensive; the efficiency and the quality that a user acquires knowledge are obviously improved, and the time of the user is greatly saved.

The invention discloses a pre-fetching-based phishing web page detection method, and relates to the acquisition of website information and the extraction and classification of topological characteristics and mainly aims to solve problems on phishing web page detection capacity. In the method, a user interface module 1 serves as an interface, a master control module 2 serves as a center, and a classifier module 3, a characteristic extraction module 4 and a web page extraction module 5 are scheduled, wherein the classifier module needs training in a training set and adopts an incremental updating mode to ensure that a classifier keeps capacity in the detection of new phishing web pages; the characteristic extraction module mainly extracts the pre-fetched characteristics of topological website structures, saves the characteristics into a training set database and simultaneously transmits the characteristics to the classifier module; and the web page extraction module captures a certain number of web pages of a given website according to an instruction of the master control module and saves the web pages into a web page database. Through the pre-fetching-based phishing web page detection method provided by the invention, both accuracy and recall rate are greatly improved.

The invention relates to a natural scene text detection method based on an attention mechanism convolutional neural network. Firstly, image data with text under natural scene is labeled and divided into training set and test set. Then the text image is processed as training data by data enhancement method. Based on the attention mechanism module and Inception network, a feature extraction networkis constructed, and the features of different fonts are learned by multi-scale feature fusion. The network is pre-trained using course learning strategies. The network is trained again by using the text image data of the natural scene. The fused features are adopted to regress to get the coordinates of the text in the image, and the text detection results are obtained. Finally, the validity of thetrained neural network is verified in the test set and other open data sets. The method can solve the problems of low character recall rate and low recognition accuracy under complex natural environment in the current technology, and has advantages in running speed.

The embodiment of the invention discloses a video clip detecting method and device, and belongs to the field of image processing. The video clip detecting method comprises the steps that a currently-input video file or real-time streaming video is used as an object to be detected, whether video in a video library appears in the current video stream or not is detected and recognized through the video analysis and retrieval method, the accurate position, in the stream to be detected, of a video clip is located and a detection result is output. Image features extracted through the video clip detecting method and device integrate image global information and image local information, the matching error rate is reduced while the high recall rate is guaranteed, robustness is kept for image content changes, therefore, accurate locating is conducted on video clips in real-time or off-line television programs or video programs and accurate detection and recognition of the video clips are kept on the conditions that frames of video streams are lost, mosaic blocks appear in the video streams, video resolution ratios and video frame rates are different and video contents change.

The invention discloses an industry classification method and system for text publishing. The method comprises the following steps: conducting first level industry category feature word collection excavating and second level industry category module training, and establishing a two-level hierarchical category system composed of the first level industry category feature word collection and the second level industry category feature word collection; conducting first level industry and second level industry classification to the published text according to the two-level hierarchical category system. The method and system provided by the invention can improve the accuracy and recall rate of the industry category of the published text.

The invention provides a quick traffic signal lamp detection method based on depth characteristic learning, and relates to the fields of image processing, deep learning and intelligent traffic. The method comprises the steps: extracting a traffic signal lamp candidate region from a detected image; and carrying out the classification of the traffic signal lamp candidate region through a convolutionneural network. The adding of the training data can enable a network to apply to various types of complex scenes, thereby improving the recall rate of traffic signal lamps and the detection accuracy.Because a traffic signal lamp candidate region extraction algorithm and a classification network can achieve the higher recall rate and the classification accuracy, the classification network is enabled to adapt to various types of complex scenes. The method is high in detection rate, and meets the real-time requirements of an unmanned vehicle. The number of candidate regions is reduced, the subsequent calculation burden of the classification network is reduced, and the whole detection rate of a system is reduced. The method can be suitable for the traffic signal lamps in various complex scenes, and improves the detection accuracy.

The invention provides a fine-grained word representation model-based sequence labeling model, which is used for performing a sequence labeling task, and belongs to the field of computer application and natural language processing. The structure of the model is mainly composed of three parts including a feature representation layer, a BiLSTM layer and a CRF layer. When the sequence labeling task is performed by utilizing the model, firstly an attention mechanism-based character level word representation model Finger is proposed for fusing morphological information and character information ofwords; secondly the Finger and a BiLSTM-CRF model finish the sequence labeling task jointly; and finally a result with F1 of 91.09% is obtained in a CoNLL 2003 data set in end-to-end and no any feature engineering forms by a method. An experiment shows that the designed Finger model remarkably improves the recall rate of a sequence labeling system, so that the model identification capability is remarkably improved.

The invention discloses a Chinese automatic answering method and a system. The invention solves the problems of huge work quantity and narrow cover range for establishing a Chinese question and answer statement base in the prior art when the Chinese automatic answering is achieved by a mode matching technique. The method comprises the steps that: the style and the key word sequence of the Chinese statement input by customer are determined; a key word sequence that is the same as the style of the statement input by the customer is searched from the question and answer statement base; the key word sequence of the statement input by the customer is matched with the searched key word sequence; and the answering statement corresponding to the key word sequence which is matched successfully is returned to the customer. The system comprises a customer statement analyzing module and an inquiring module. The invention has the advantages of increasing the content cover range of the question and answer statement base, reducing the work load for establishing the base, and obviously enhancing the accuracy and the recall rate of the Chinese automatic answering system.

The invention discloses a method for automatically creating a keyword index table. The method includes subjecting a file to be translated to word segmentation process to obtain a word list of the file, and subjecting the word list to part-of-speech tagging; filtering candidate keywords in the word list to obtain a coarse candidate word collection and codes of each sense of the candidate keywords; subjecting the candidate keywords to synonym chain construction according to semantic similarity of the words to obtain a synonym chain collection; acquiring word weight of vocabularies in the synonym chain collection and extracting keywords to form a keyword collection according to the word weight; and comparing the keyword collection with an existing reference library keyword index collection, providing a relevant file collection if the existing reference library keyword index collection contains the candidate keywords, otherwise, adding the candidate keywords to the reference library keyword index collection, and simultaneously, creating an index. Compared with traditional keyword extraction methods, the method has the advantages that precision rates and recall rates are obviously increased.

The invention relates to a medical search engine and a system based on a repository. The engine works as follows: capturing a Chinese medical health directory to establish an original medical webpage database, extracting related information on webpage in the original medical webpage database and extracting comment information on hospitals, departments and doctors, so as to establish a medical comment information database, carrying out medical comment attribute field extraction of the abstracted related information by means of term frequency statistics and questionnaire to extract viewpoint phrase, analyzing viewpoint phrase orientation, determining an analytic result showing whether the comment information is positive or negative, determining the ranking of hospitals, departments and doctors, ordering search results according to a medical repository, and providing a user with highly structured and highly related information. In order to overcome the disadvantages of the result information of a common search engine such as unstructured form and low correlation degree and accuracy, the medical search engine and the system establish the medical repository to provide a user with highly structured medical information, and increase both correlation degree and accuracy for the user during querying medical information; moreover, the medical search engine and the system can effectively increase the accuracy and the recall rate of search results.

The invention puts forward a synonym expansion method and device for search information. The method comprises the following steps that: carrying out word segmentation processing on the search information to obtain at least one segmented word of the search information; obtaining the candidate synonym set of the segmented word, wherein the candidate synonym set comprises at least one synonym of thesegmented word; forming a synonym pair which contains the synonym and the segmented word by aiming at each synonym; carrying out feature extraction on the synonym pair to obtain a synonym pair featureset; according to the feature set, predicting the synonym pair to obtain a target probability that the synonym pair is predicted as reasonable replacement; and if the target probability exceeds a preset threshold value, forming a synonym expansion item by the segmented word and the synonym, and searching and obtaining a search result on the basis of the synonym expansion item. Through the method,synonym replacement rationality and accuracy can be improved, the recall rate and the accuracy of a search result are improved, and therefore, the technical problems in the prior art that synonym replacement is inaccurate and a search result recall rate is poor can be solved.

The present invention relates to network computing and in particular to a method and respective system for operating a Web Services search method in a networked environment—in particular the Internet, wherein a Service Broker is requested by a Service Consumer for offering Web Services, and a Service Provider provides a requested Web Service to the Service Consumer, and wherein a Service Provider registers Web Services at the Service Broker and wherein the Web Services are implemented preferably by respective Remote Portlets running on a Portal Server of the Service Provider. In order to increase the search precision and the recall rate it is proposed to perform the steps of: a) retrieving data mining input data relating to description details of the inquired Web Services from respective Service descriptions of the inquired Web Services, b) performing a data mining function on said data mining input data, wherein a clustering is performed with a distance calculation function including said description details and wherein the clustering yields a cluster model comprising a plurality of clusters and a mapping for each Web Service to one of said clusters, wherein Web Services having a similar semantic meaning are collected in a single cluster, yielding a data mining result, c) adding cluster information comprised of said data mining result to a service model and storing this cluster information in a database, d) offering search response data based on said cluster model and service model.

The invention provides a Chinese short text classification method based on characteristic extension, and the method comprises the following steps that (1) a background knowledge base is established: the two-tuples of feature words which meet a certain constraint condition are dug from a long text corpus with category marks to form the background knowledge base; (2) short text which is trained in a centralized way is extended: extension words are added to the short text which is trained in a centralized way according to a certain extension rule according to the two-tuples in the background knowledge base; (3) a classification model is built: a (shared virtual memory) SVM classification model is established through an extended short text training set; (4) the short text to be classified is extended: the extension words are added to the short text to be classified according to a certain extension rule according to the two-tuples in the background knowledge base and the feature space of the classification model; and (5) a classification result is generated: the classification result is generated through the classification model and the extended short text. According to the Chinese short text classification method based on characteristic extension, the features of the short text are enriched through the long text corpus, so that the accuracy and the recall rate in the classification of the short text are improved.

The invention discloses a recognition method and device for application search intentions and an application search method and server. The method includes the steps that search terms in query sessions are acquired from a query session log of an application search engine; label systems of the search terms are mined according to the search terms in the query sessions and preset strategies; the application search intentions corresponding to the search terms are recognized according to the label systems of the search terms. In the scheme, by proposing the label method of the recognition method for the user intentions corresponding to the app label systems, the fine-grained user query intentions are flexibly expressed. The label systems of the user intentions are constructed based on the unsupervised machine learning technology, a traditional user intention classification method is abandoned, an automatic user intention mining process is achieved, and a high-accuracy and high-recall-rate user intention label list can be generated; the user intentions and apps are mapped into the same label system, and thus a user can rapidly and accurately acquire apps meeting intentions when searching for apps.

The invention discloses a course recommendation method based on students' attributes. The course recommendation method comprises the following steps: S1, calculating the similarity between target students and other students according to the students' attributes and finding out a 'neighbor' student collection similar to the target students in elective hobby; S2, finding out courses which are not elected by the target students but elected by 'neighbor' students to recommend. According to the course recommendation method, based on students' attribute data, more accurate course recommendation can be realized, and meanwhile, higher recall rate and coverage rate are obtained.

The invention relates to an answer extraction method for a tourism question and answer system based on ontology reasoning, which belongs to the field of artificial intelligence. The answer extraction method for the tourism question and answer system based on ontology reasoning is characterized in that: firstly, semantic rules in the field are defined, an artificial ontology knowledge base is constructed, and question sentences of users are analyzed; secondly, answer extraction is performed by combing semantic rule-based reasoning and information retrieval, instead of simple matching; and thirdly, corresponding answer extraction algorithms are designed according to different question sentence types. The invention provides the answer extraction method for the tourism question and answer system based on ontology reasoning, which introduces the ontology concept into construction of the knowledge base of the question and answer system, uses an OWL (Ontology Web Language) ontology descriptive language to clearly and definitely represent concepts, attributes and relations in the field of tourism, and more effectively organizes knowledge. In an open test, the precision of answers of the question and answer system based on ontology reasoning to 1346 natural language questions of the users reaches 81.35 percent, and the recall rate reaches 90.49 percent.

The invention discloses a CRFs (conditional random fields) and SVM (support vector machine) based method for extracting fine-granularity sentiment elements in product reviews. The method comprises the steps as follows: a, a CRFs model is adopted, review language characteristics are taken as sequences, then position labelling is performed on review languages according to the sequences, corresponding rules are adopted to perform stratified filtering on error labels, and extraction for sentiment subjects and sentiment words is finished; and b, an SVM model is adopted to perform sentiment orientation analysis on word pairs according to the extracted sentiment subjects and sentiment words as well as introduced sentence structure features. According to the invention, the sentiment subjects and the sentiment words in review sentences are extracted together, further, sentiment classification accuracy in the sentiment orientation analysis is improved, so that the sentiment element extraction and sentiment judgment are improved, and F value is up to 76.3%; due to introduction of word meaning codes, the generalization ability and the robustness of a system are improved by virtue of the word meaning codes, and the accurate rate and recall rate of review result analysis are greatly improved.

The invention discloses a method for fusing a visible image and an infrared image. According to the method, expanded foreground of an infrared domain is used to act as an initial mask for visible light and a gray scale image, then infrared foreground and mask visible light foreground are fused so as to realize composition of complementary regions from different domains, mask processing is carried out on the infrared foreground by using the fused information, connected regions in a foreground image acquired from results are extracted out to act as candidate targets, and finally, an active contour model is applied to each of the connected regions in the infrared domain so as to detect the boundary of the target. According to the method, complementary information in the visible image and the infrared image are extracted to realize fusion of the images. Compared with other methods, a better recall rate can be acquired.

The present invention provides an information recommendation method and apparatus. The method comprises: after n question and answer interactions with a user, receiving a current question of the user, and extracting a core word corresponding to each question of the user, so as to form a core word sequence corresponding to a user question sequence; searching a knowledge base for a candidate question set corresponding to the current question; searching an associated question sequence information base for a target associated question sequence corresponding to the core word sequence, and from the target associated question sequence, extracting a current associated question corresponding to the current question; calculating a similarity value between each question in the candidate question set and the current associated question and a similarity value between the current associated question and the current question separately; and when a calculated greatest similarity value is greater than a similarity threshold, providing the user with answer information of a question corresponding to the greatest similarity value; or when the greatest similarity value is lesser than the similarity threshold, providing answer information of the current associated question. According to the scheme provided by the present invention, the accuracy and the recall rate of the answer information recommended to users are improved.

The embodiment of the invention provides a face clustering method, device and system and a storage medium. The method includes: acquiring a plurality of face images; detecting face quality of target faces in the plurality of face images to obtain face quality data of the plurality of face images; extracting features of target faces in at least partial face images in the plurality of face images toobtain face feature data of the at least partial face images; and carrying out clustering on the at least partial face images according to the face feature data and the face quality data of the at least partial face images. According to the face clustering method, device and system and the storage medium of the embodiment of the invention, the face features are considered in clustering, the facequality is also considered, and impacts of poor face quality or large face quality difference on a clustering effect are enabled to be effectively reduced in a clustering process. The face clusteringmethod has the characteristics of a high accuracy rate, a high recall rate, high reliability and the like.