369 results about "Categorical models" patented technology

The present invention provides a method for recognizing a named entity included in natural language, comprising the steps of: performing gradual parsing model training with the natural language to obtain a classification model; performing gradual parsing and recognition according to the obtained classification model to obtain information on positions and types of candidate named entities; performing a refusal recognition process for the candidate named entities; and generating a candidate named entity lattice from the refusal-recognition-processed candidate named entities, and searching for a optimal path. The present invention uses a one-class classifier to score or evaluate these results to obtain the most reliable beginning and end borders of the named entities on the basis of the forward and backward parsing and recognizing results obtained only by using the local features.

A method for building a classification model for classifying unclassified documents based on the classification of a plurality of documents which respectively have been classified as belonging to one of a plurality of classes, said documents being digitally represented in a computer, said documents respectively comprising a plurality of terms which respectively comprise one or more symbols of a finite set of symbols, and said method comprising the following steps: representing each of said plurality of documents by a vector of n dimensions, said n dimensions forming a vector space, whereas the value of each dimension of said vector corresponds to the frequency of occurrence of a certain term in the document corresponding to said vector, so that said n dimensions span up a vector space; representing the classification of said already classified documents into classes by separating said vector space into a plurality of subspaces by one or more hyperplanes, such that each subspace comprises one or more documents as represented by their corresponding vectors in said vector space, so that said each subspace corresponds to a class.

The invention discloses a Chinese event extraction method. The method comprises the following steps of: 1) carrying out entity identification on a to-be-extracted Chinese text, taking identified entities as candidate words of event elements, labeling the Chinese text word by word, and identifying an event triggering word in the Chinese text and event types described by the Chinese text by combing a labeling mode according to the labeling result; 2) inputting the Chinese text, the candidate words of the event elements, the event triggering word and the event types into a classification model, and judging whether each candidate word is a real event element or not; and 3) obtaining a whole event structure according to the obtained event elements, the event triggering word and the event types so as to complete event extraction. According to the method disclosed by the invention, the problem that the event triggering word is not completely matched with the words in the text in the Chinese event extraction is solved, and the correctness of the Chinese event extraction is improved.

The present invention relates to a method of detecting anomalies in signaling flows in a communication device connected to a database. In accordance with the method, a communication device receives (301) labeled learning signaling flows and feeds these flows to the database, the signaling flows being labeled to either normal signaling flows or to different signaling flows indicative of attacks. Then a profile specific classification model is built (307) by using the learning signaling flows contained in the database, the profile being a model that characterizes a signaling flow that corresponds to either a packet, transaction or dialog. Next the learning signaling flows are classified (309), the signaling flows being classified to either normal signaling flows or to different signaling flows indicative of attacks, the classification being based on the classification model. Then a new signaling flow is received (317) and at least one attribute is extracted from the received signaling flow, and by using the at least one extracted (319) attribute for the received signaling flow is classified either to a normal signaling flow or to a signaling flow indicative of an attack, the classification being based on the classification model.

The invention provides a method of constructing a text classification model. The method comprises the steps of: constructing a training sample set according to structure features of characters, wordsand sentences of text information, wherein each piece of sample data in the training sample set corresponds to a feature matrix A of a piece of text information about the words and a feature matrix Babout the characters and a category vector O corresponding to the piece of text information, and the dimension number of O is the same as a category number; and using the feature matrix A about the words and the feature matrix B about the characters in the training matrix set as input and the corresponding category vector O as output to train a deep learning model to obtain the text classificationmodel. Classification is carried out according to the classification model constructed by the method, an accuracy rate of text classification can be improved, and the method is particularly suitablefor use in short-text classification.

A Markov Random Field (MRF)-based technique is described for performing clustering of images characterized by poor or limited data. The proposed method is a statistical classification model that labels the image pixels based on the description of their statistical and contextual information. Apart from evaluating the pixel statistics that originate from the definition of the K-means clustering scheme, the model expands the analysis by the description of the spatial dependence between pixels and their labels (context), hence leading to the reduction of the inhomogeneity of the segmentation output with respect to the result of pure K-means clustering.

The present invention provides an information recommendation method and apparatus in social media. The method comprises an offline procedure and an online procedure. The offline procedure comprises: determining a point of interest (POI) of a target user from historical operation information of the target user in social media; selecting, from historical operation information of high-quality users related to the POI in the social media, information related to the POI for use as a labeled corpus; and training an interest classification model of the target user by using the labeled corpus as a training sample. The online procedure comprises: acquiring to-be-recommended information of social media; inputting the to-be-recommended information to the interest classification model of the target user, so as to determine whether the to-be-recommended information conforms to the interest of the target user; and if the to-be-recommended information conforms to the interest of the target user, recommending the to-be-recommended information to the target user. By means of the present invention, the efficiency of information recommendation in social media can be improved.

The application provides a method and device for recognizing a sentence. The method comprises the following steps: determining a non-stop word as a key word for an acquired sentence to be recognized; selecting a candidate sentence including the key word of the sentence to be recognized in a preset sentence library; determining a subject classification tag and an intention classification tag for the sentence to be recognized through a classification model which is built in advance, wherein the classification model can recognize an intention of an unknown type; when the recognized intention classification tag is of an unknown type and a plurality of candidate sentences exist, grouping the candidate sentences according to a preset intention tag; displaying the preset information corresponding to the candidate sentence in each group. Different groups correspond to different intention types, and the candidate sentence in each group is selected as a target sentence to display the preset information corresponding to each target sentence, thus, the problem that information fed back is single or even cannot be fed back is solved.

Embodiments of the invention relate to improvements to the support vector machine (SVM) classification model. When text data is significantly unbalanced (i.e., positive and negative labeled data are in disproportion), the classification quality of standard SVM deteriorates. Embodiments of the invention are directed to a weighted proximal SVM (WPSVM) model that achieves substantially the same accuracy as the traditional SVM model while requiring significantly less computational time. A weighted proximal SVM (WPSVM) model in accordance with embodiments of the invention may include a weight for each training error and a method for estimating the weights, which automatically solves the unbalanced data problem. And, instead of solving the optimization problem via the KKT (Karush-Kuhn-Tucker) conditions and the Sherman-Morrison-Woodbury formula, embodiments of the invention use an iterative algorithm to solve an unconstrained optimization problem, which makes WPSVM suitable for classifying relatively high dimensional data.

The invention discloses a customizable voice wake-up method and a system. Through using a long and short memory network and a connection time sequence classification model, phoneme information of the voice information is modeled, the model is trained, the model after training is adopted for testing, and a possible phoneme sequence most similar to a customized wake-up word is searched in a generated Lattice network structure and serves as a judgment basis. The feature of CTC model output posteriori probability sparsity is used for high-efficiency searching, and the technique of calculating the confidence of the wake-up word is completed. On one hand, high wake-up performance can be obtained, that is, high accuracy and low error wake-up can be obtained; and on the other hand, the calculation resources of the application system are relatively little consumed.

A method for learning a classification model by a computer system is disclosed. One or more positive class data and one or more negative class data are prepared. The classification model is trained based on the positive class data to adjust one or more parameters of the classification model so that the positive class data is reconstructed by the classification model. The classification model is trained based on the negative class data to adjust the one or more parameters so that the negative class data is prevented from being reconstructed by the classification model. For the negative class data, changes in the one or more parameters with gradient of an objective function may be calculated using an unsupervised learning algorithm. The one or more parameters may be updated based on the changes in an opposite manner to the training based on the positive class.

The invention relates to a multi-feature fusion phishing webpage detection method, which comprises two parts such as a training process and a detection process. The multi-feature fusion phishing webpage detection method integrates three views of phishing webpage characteristics by combining a semi-supervised learning tri-training method, and mainly solves a problem that the existing phishing webpage detection methods mostly need to perform classification model training by using supervised learning through a large amount of annotation data. The method provided by the invention mainly combines a coordinated training algorithm, starts from webpage URL characteristics, webpage information characteristics and webpage search information characteristics, applies the idea of multiple views and multiple classifiers to phishing webpage detection, and achieves the purposes of reducing the total numbers of manual annotation training samples and timely recognizing a phishing webpage through coordinated training and learning of different classifiers.

The invention relates to a method for classifying field questions by integrating with syntax structure relationship and field characteristics, which comprises the following steps: field terms are collected; a special field question classification system is defined; the syntax structures of the questions are analyzed; a sentence trunk is extracted; the sentence truck and the field vocabulary are taken as question classification characteristics; a question classification model is built through an improved Bayesian classification algorithm method; a special field question classification training corpus and a test corpus are set up; a special field question classifier is set up. Because question classification is a very important step in an answering system and a key factor for establishing answer extraction strategies and positioning answers, the method of the invention can select the sentence trunk and the field vocabulary as the classification characteristics based on the syntax structure analysis by integrating with the field characteristics, adopts the improved Bayesian classification algorithm method, builds the question classification model and takes the question classification test in the field of Yunnan tourism; the result shows the method is effective, and improves the field question classification accuracy, thereby offering consultancy service to the users with high efficiency, quickly and accurately.

The invention discloses a text classification method based on feature information of characters and terms. The method comprises the steps that a neural network model is utilized to perform character and term vector joint pre-training, and initial term vector expression of the terms and initial character vector expression of Chinese characters are obtained; a short text is expressed to be a matrixcomposed of term vectors of all terms in the short text, a convolutional neural network is utilized to perform feature extraction, and term layer features are obtained; the short text is expressed tobe a matrix composed of character vectors of all Chinese characters in the short text, the convolutional neural network is utilized to perform feature extraction, and Chinese character layer featuresare obtained; the term layer features and the Chinese character layer features are connected, and feature vector expression of the short text is obtained; and a full-connection layer is utilized to classify the short text, a stochastic gradient descent method is adopted to perform model training, and a classification model is obtained. Through the method, character expression features and term expression features can be extracted, the problem that the short text has insufficient semantic information is relieved, the semantic information of the short text is fully mined, and classification of the short text is more accurate.

The invention discloses an internet public opinion analysis method. The internet public opinion analysis method comprises the steps of firstly for selected and acquired events, partitioning a source text of a microblog and removing partition items unrelated to sentiments; secondly making statistics by adopting a statistic analysis tool to obtain an input of a sentiment classification model; and finally for the input, modeling related words, expressions and symbols capable of expressing the sentiments in the microblog content by using a classification algorithm, giving out comprehensive sentiment index assessment, obtaining sentiment categories, and performing public opinion monitoring and sentiment trend analysis. According to the method, the words, the expressions, the symbols and the like in the microblog are subjected to sentiment modeling, and the response situations of hot events in the microblog can be automatically classified and effectively monitored through sentiment index calculation, so that the public opinion risk can be effectively assessed and intemperate events can be prevented and controlled.

The invention discloses an abnormality detecting method and device and relates to the technical field of abnormality detection. The method comprises acquiring target log data to be detected; acquiring a first probability that the target log data belong to the category of abnormality through preset categorization models which are acquired by training a plurality of access sample data with categories determined; determining whether the first probability is higher than a preset threshold value; if so, determining abnormality of the target log data. Therefore, the abnormality detecting method solves the technical problem such as large data labelling volume, high labelling costs, sensitive parameters and linear inseparability in existing abnormality detecting methods and has the advantages of reducing data labelling volume and labelling costs and improving detecting performance.

The invention provides a Python program type defect detection method based on abstract syntax trees. The method comprises the steps of (1) collecting Python software defect report information and extracting a defect identifier and a defect error type, (2) obtaining source codes of two versions of programs before and after defect repair, (3) generating abstract syntax trees corresponding to the twoversions of source codes, matching and obtaining a change function node and marking the defect error type, (4) generating a feature vector of a defect code according to the context information of thechange function node, (5) training a multi-classification models on a defect code feature vector by using the machine learning technology, and (6) reminding a developers of possible type defect information in testing a Python program file. The invention aims to solve the problems of the lack of the type defect analysis for a Python language and the inability to detect a possible defect error typeat present, thus the management of software quality is guided, and the maintainability of the software is improved.

The present invention pertains to a system and method for the tokenization of text. The featurizer may be configured to receive input text and convert the input text into tokens. According to one aspect of the invention, the tokens may include only one type of character, the characters selected from the group consisting of letters, numbers, and punctuation. The tokenizer may also include a classifier. The classifier may be configured to receive the tokens from the featurizer. Furthermore, the classifier may be configured to analyze the tokens received from the featurizer to determine if the tokens may be input into a predetermined classification model using a preclassifier. If one of the tokens passes the preclassifier, then the token is classified using the predetermined classification model. Additionally, according to a first aspect of the invention, the tokenizer may also include a finalizer. The finalizer may be configured to receive the tokens and may be configured to produce a final output.

The embodiment of the specification discloses a processing method, a device and a system for predicting transaction risk. The method comprises the following steps: acquiring real-time transaction data, extracting features from the real-time transaction data, and obtaining a first feature set; the first feature set is classified by using a constructed classification model to obtain a classificationresult, the classification model comprising a model obtained by training based on the marked transaction data in the historical transaction data; detecting outlier of the first feature set by usingthe constructed description model, and obtaining the detection result, the description model comprising a model generated by clustering processing based on historical transaction data; determining a risk prediction result of the real-time transaction according to the classification result and the detection result. By utilizing the various embodiment of the present specification, it is possible toincrease the probability of predicting the risk of a transaction or a user, and at the same time to identify emerging means of committing a crime, thereby preventing from being bypassed by an unlawfulelement.

Provided are a method and system for modeling a service using a Bayesian network and status information in a distributed environment. The method includes creating a scenario for modeling at least one service, categorizing the service into models according to properties based on the scenario, setting interrelationships including a chronological relationship, a hierarchical relationship, and correlation between the categorized models, and deriving information including a service goal and status information with respect to the models and completing service modeling. Accordingly, a service provider, which digitally provides various services in a ubiquitous environment, can efficiently and actively provide intelligent services. In addition, since a specific method of modeling a service based on probabilities is provided, an intelligent service is modeled efficiently and diversely.

The invention discloses an image segmentation method and device and belongs to the field of image processing. The image segmentation method includes the following steps: establishing a significance model of an image; according to the significance model, obtaining foreground sample points and background sample points in the image; according to the significance model and the foreground sample points and the background sample points, establishing a foreground and background classification model; and according to a preset image segmentation algorithm, segmenting the image, wherein the preset image segmentation algorithm uses the foreground and background classification model and edge information between pixel points to segment the image. Through automatic determination of the foreground and background sample points in combination with the significance model, the foreground and background classification model is established and the foreground and background classification model is used to realize image segmentation. Therefore, problems, which exist in related technologies, that the foreground sample points and the background sample points must be selected roughly manually and the segmentation efficiency is comparatively low when a large quantity of images are segmented are solved so that effects of realizing automation selection of samples and improving the classification precision and segmentation efficiency are achieved.

The present invention is directed towards systems and methods for the classification or scoring of content items. The method according to one embodiment comprises providing at least one labeled content item as input to an initial classification model, a given labeled item having at least one feature, a given feature being associated with a weight, computing a posterior probability of the initial classification model for the given labeled content item and generating an updated classification model using the initial classification model and the weight associated with the given feature in the given labeled content item. The updated classification model is applied to an unlabeled content item to determine a score for the unlabeled content item.

The invention discloses a short text garbage identification and modeling method and device. The short text garbage identification and modeling method includes the steps that word segmentation is conducted on a short text to be determined, word sets are acquired, and garbage features of the short text to be determined are analyzed to acquire analytical information; the analytical information of the short text to be determined and each word in the word sets are compared with feature elements in predetermined feature element sets respectively, and word feature vectors of the short text to be determined are generated according to feature values of words or the analytical information matched with the feature elements in the feature element sets; whether the short text to be determined is a garbage text or not is determined according to the word feature vectors of the short text to be determined and classification models; the classification models are trained in advance, wherein the classification models combine the number of samples with centralized training and select a proper classification algorithm. Due to the fact that the word feature vectors of the feature values of the analytical information are expanded to conduct garbage identification, the identified accuracy rate for identifying the garbage texts is improved.

A closed loop system is presented for selecting samples for labeling so that they can be used to generate classifiers. The sampling is done in phases. In each phase a subset of samples are chosen using information collected in previous phases and the classification model that has been generated up to that point. The total number of samples and the number of phases can be chosen by the user.

The invention provides a telecommunication fraud event detection method and a telecommunication fraud event detection system. The constructed telecommunication fraud event detection model is used forpredicting a fraud mode. The construction of the fraud event detection model comprises the following steps: establishing a dynamic communication graph reflecting different time series communication modes based on existing fraud behavior data; an abnormal sub-graph sequence with fraud behaviors is mined based on the dynamic communication graph, and the abnormal sub-graph sequence is composed of abnormal nodes, nodes associated with the abnormal nodes and edges; and training a multi-classification model based on the abnormal subgraph sequence, and obtaining a telecommunication fraud event detection model for detecting multiple fraud modes. By utilizing the detection method and the detection system, the fraud behavior can be quickly and accurately predicted.

The invention discloses a method for automatically identifying and classifying rock lithology in a deep learning mode, aiming at analyzing rock lithology in geological engineering. The method includesthe following steps: A. based on the kinds of rocks, acquiring rock images of different types, and grouping the rock images into a training group and a test group; B. taking a convolutional neural network Inception-v3 model as a pre-training model, acquiring image features by using a feature extraction model of the pre-training model; C. establishing a Softmax regression model; D. training a model for automatically identifying and classifying rock images; and E. testing the model for automatically identifying and classifying rock images. According to the invention, the method herein, by establishing the model for automatically identifying and classifying rock images, can analyze the geological conditions in an automatic and intelligent manner, greatly save labor and materials, and reducescost.

The present application relates to a machine learning technology in the field of artificial intelligence and provides a medical record data processing method, a medical record data processing apparatus, a computer device and a storage medium. The method includes the following steps that: feature words are extracted from source medical records, so that a feature word set is obtained, wherein the feature word set includes identity feature words and pathological feature words; and the matching degree of the feature word set and a feature word set corresponding to historical medical records in a historical medical record set is calculated, reference medical records are selected according to calculation results, and a reference medical record set is obtained; a trained classification model is used to obtain the patient categories of patients corresponding to the source medical records according to the identity feature words; screening factors corresponding to the patient categories are obtained, and the reference medical records in the reference medical record set are sequenced according to the screening factors, and a first predetermined number of reference medical records are selectedas target reference medical records according to the sequencing results of the reference medical records; and the target reference medical records are sent to doctor terminals. With the method adopted, the efficiency and accuracy of clinical assistant decision-making can be improved.

The invention discloses a text feature extraction method based on categorical distribution probability. The text feature extraction method based on the categorical distribution probability extracts text feature words by means of the manner according to which categorical distribution difference estimation is carried out on words of a text to be categorized. Mean square error values of probability distribution of each word at different categories are worked out by means of category word frequency probability of the words. A certain number of words with high mean square error values are extracted to form a final feature set. The obtained feature set is used as feature words of a text categorizing task to build a vector space model in practical application. A designated categorizer is used for training and obtaining a final category model to categorize the text to be categorized. According to the text feature extraction method based on the categorical distribution probability, category distribution of the words is accurately measured in a probability statistics manner. Category values of the words are estimated in a mean square error manner so as to accurately select features of the text. As far as the text categorizing task is concerned, a text categorizing effect of balanced linguistic data and non-balanced linguistic data is obviously improved.

The invention discloses a discount coupon issuing method and system. The discount coupon issuing method comprises the following steps of S1, obtaining user historical data, wherein the user historical data comprises user browsed information data, user portrait data and a user order-placing mark, and the user order-placing mark is used for representing whether a user places an order at a historical check-in day or not; S2, processing the user historical data through a machine learning algorithm, and building a classification model; S3, predicting an order-placing probability of the user by utilizing the classification model; and S4, issuing a discount coupon according to the order-placing probability of the user. According to the method and the system, whether the discount coupon needs to be issued or not is judged by predicting the order-placing probability of the user through the machine learning algorithm, and the discount coupon is pointedly issued to the user with the relatively low order-placing probability, so that the operation efficiency of discount coupons is improved and the increment of an order quantity can be better stimulated.