A method and system for extracting text sentiment reasons based on clustering multi-kernel learning
By applying a multi-core learning model based on clustering perspective in the field of mental health assessment, the shortcomings of the emotional cause extraction method in the existing technology are solved, and more accurate text emotional cause extraction and psychological crisis warning analysis are achieved.
Patent Information
- Application Number
- CN202311281213.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-28
- Publication Date
- 2025-04-04
AI Technical Summary
The existing emotional cause extraction methods have problems such as time-consuming construction of a rule base, complex feature engineering, large amount of deep learning calculations and limited generalization capabilities, and it is difficult to achieve accurate text emotional cause extraction and psychological crisis warning analysis in the field of mental health assessment.
A multi-core learning model based on the clustering perspective is adopted, and the clustered objective function is used as the classification objective function, and a model is constructed in a high-dimensional kernel space. By combining information extraction and relationship extraction technologies, multi-classification processing and extraction of emotional causes is realized.
It has achieved a more accurate text emotional reasons extraction in the field of mental health assessment, made accurate early warning analysis for psychological crises, overcome the shortcomings of the existing technology, and had the characteristics of efficiency, accuracy and strong adaptability.
Smart Images
Figure CN119760130A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of psychology, and more specifically, relates to a text emotion cause extraction method and system based on clustering multi-kernel learning for mental health assessment, which observes and solves classification problems from a clustering perspective and combines the advantages of the two. Background Art
[0002] As social competition becomes increasingly fierce and interpersonal relationships become increasingly complex, more and more people are under various pressures and are prone to psychological problems such as anxiety and depression. In March 2022, Dingxiang Doctor released the "2022 National Health Insight Report", which reviewed and summarized the national health status in 2021 based on data insights from 60,420 samples in multiple dimensions such as self-assessment of current status, self-assessment of expectations, and health behaviors. Among them, there are many contents and data involving mental health, from which we can see the degree of attention people pay to mental health and the changes in the public's mental health level. Mental health and physical health have become equally important to people. The ranking of attention to mental health-related knowledge has risen to the most concerned health content outside of sports and fitness, diet and nutrition knowledge, and popular science of common diseases. Not only has the attention increased, but people are also taking positive actions. The relevant report pointed out that 74% of people are taking positive actions to cope with and solve psychological distress. As professional psychological services are more widely popularized and recognized in China, more people are beginning to seek help from professional institutions. Nowadays, psychological services are no longer limited to psychological counseling. More lightweight psychological content products are more popular, and various pan-psychological content platforms and psychological counseling platforms have begun to emerge and develop.
[0003] Current mental health research uses data from various social networking platforms to automatically assess users' mental health. Researchers store social network text content in the form of documents as sample data, and use specific rules for evaluation. By extracting emotional causes from the samples to be tested, the purpose of mental health assessment and early warning of mental illness can be achieved. In the field of psychology, emotion cause extraction refers to the emotion words (or emotion clauses) and documents of a given document, where the document consists of several clauses, each of which consists of several words. The goal of the emotion cause extraction task is to extract the cause clause corresponding to the emotion.
[0004] Nowadays, there are three main types of sentiment cause extraction methods: rule-based methods, feature-based methods, and deep learning-based methods. Rule-based methods establish rules for language analysis or generation based on linguistic theory, describe the various components of language and the structural and semantic relationships between the components, and use these rules for analysis. Feature-based methods mainly classify sentiment causes by extracting features from texts; feature-based methods usually include the following steps: feature extraction, extracting sentiment-related features from texts, which can be information on vocabulary, syntax, semantics, etc. Common features include word frequency, part of speech, sentiment dictionary matching, etc. Feature representation, converting the extracted features into a form that can be processed by machine learning algorithms. Common representation methods include vectorization and unique hot encoding; classification model training, using machine learning algorithms to train the extracted features and build a sentiment cause classification model. Common classification algorithms include support vector machines (SVM), naive Bayes, random forests, etc.; sentiment classification, using the trained model to classify the sentiment causes of test sample texts. Deep learning-based methods break through the limitations of hardware and software, allowing researchers to use deep neural networks to solve the problem of sentiment cause extraction.
[0005] The above three types of emotion cause extraction methods all have their own advantages and disadvantages.
[0006] 1. The rule-based method is a traditional method for extracting emotional causes. It requires manual writing of rules. The advantages of this method are simplicity and effectiveness, but the disadvantages are also obvious:
[0007] (1) Building a high-quality rule base requires a lot of manpower and time. Moreover, the rules are based on experts, which is subjective and difficult to be complete.
[0008] (2) When the rules are changed or updated, the maintenance cost is high (manual summary of the rules is required);
[0009] (3) When there are too many rules, conflicts may occur between rules and other rules, making them difficult to maintain. Finally, it is difficult to expand to other scenarios and often requires rewriting the rules. Its performance depends on the quality and quantity of rule formulation. It is more applicable when there is less data, but there is a performance bottleneck when processing large-scale data.
[0010] 2. The feature-based method is also a common method for extracting emotional reasons. It has the following main disadvantages:
[0011] (1) Feature engineering is difficult. Feature-based methods require manual selection and design of features, which requires domain knowledge and professional experience. When dealing with different types of data, it is necessary to select appropriate features for different problems, which may make feature engineering complicated and time-consuming;
[0012] (2) Limited feature representation. Feature-based methods usually use manually designed features to represent text, which may not capture all the information in the text. For example, some important semantic information may not be represented by manually designed features;
[0013] (3) Poor generalization ability. When dealing with new fields or new types of data, feature-based methods may need to redesign and select features. This makes feature-based methods have certain limitations in generalizing to new problems;
[0014] (4) Manual selection of rules is required. Feature-based methods usually require manual selection of classification rules or thresholds for sentiment classification. This may cause the classification results to be affected by subjective factors and be difficult to adapt to data of different types and fields.
[0015] 3. Methods based on deep learning also have the following disadvantages, including:
[0016] (1) Large amount of computation. Deep learning models usually require a lot of computing resources and time to train. This may be a challenge for some devices and scenarios with limited computing power;
[0017] (2) Reliance on large amounts of data. Deep learning models usually require a large amount of labeled data for training. If the data set is small or the labeling is inaccurate, it may affect the performance of the model;
[0018] (3) Black box model. Deep learning models are often considered black box models, meaning that it is difficult to explain the decision-making process and internal mechanisms of the model. This may be a problem in some scenarios that require high interpretability.
[0019] (4) Limited generalization ability. When deep learning models process new fields or new types of data, they may need to be retrained or fine-tuned. This makes deep learning models have certain limitations in generalizing to new problems;
[0020] (5) High data requirements. Deep learning models usually require a large amount of labeled data for training. This may be difficult to obtain large-scale labeled data in some fields or tasks. Summary of the invention
[0021] Technical issues to be solved
[0022] In order to overcome the shortcomings of the above three types of emotion cause extraction methods, the present invention observes and processes the classification problem from a clustering perspective for the first time. After the samples are mapped to a high-dimensional kernel space, the objective function of clustering is used as the objective function of classification, thereby constructing a clustering-based multi-kernel learning model. Observing and solving the classification problem from a clustering perspective, combining the advantages of the two to derive a unified model, more accurately realizes the extraction of text emotion causes in the field of mental health assessment, and makes accurate early warning analysis for psychological crises.
[0023] Technical Solution
[0024] The purpose of the present invention is to provide a method and system for extracting text emotion causes based on clustering multi-core learning. By observing and processing classification problems from a clustering perspective, the objective function of clustering is used as the objective function of classification after the samples are mapped to a high-dimensional kernel space, thereby constructing a multi-core learning model based on clustering, which more accurately realizes the extraction of text emotion causes in the field of mental health assessment and makes accurate early warning analysis for psychological crises. The present invention is implemented through the following technical solutions. A method for extracting text emotion causes based on clustering multi-core learning includes the following steps:
[0025] S1. Storing psychological text data in the form of documents as a data set, where each sample in the data set is a sentence c containing emotions and reasons;
[0026] S2, define several types of sentiment reason extraction p1, p2, ..., p n ;
[0027] S3, using information extraction technology, extract the sentiment word w1 in sentence c, and extract the sentiment sentence c1 and the cause clause c2 respectively;
[0028] S4, locate the keyword w2 in the cause clause c2;
[0029] S5, after combining the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2, use the relation extraction technology to perform multi-classification processing to determine whether it meets the given sentiment cause relationship p i ;
[0030] S6, the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 meet the given sentiment cause relationship p i , the reason clause c2 is the emotional reason clause to be extracted.
[0031] The clustering multi-kernel learning model here contains a given labeled dataset
[0032]
[0033] in represents the i-th training example of the n-th class, l n is the number of training examples of the nth class,
[0034] Y={1,2,…,N}
[0035] Multi-core combination function
[0036] where k m is the base kernel function, d m is the weight, φ(x) is the characteristic function of k, and the calculation
[0037]
[0038] The clustering-based multi-core learning model is formalized as the following constrained optimization problem:
[0039]
[0040] Formula (1) is the model optimization proposed in this patent based on the clustering multi-kernel learning model and the training algorithm. m Given the problem, the solution is optimized in linear time by the stationary point method.
[0041] Furthermore, the stationary point method can be used to optimize and solve the problem in linear time as a simple linear optimization problem under the l1 norm (‖d‖1=1) constraint.
[0042] Furthermore, for the linear optimization problem under a simple l1 norm (‖d‖1=1) constraint, a nearest neighbor algorithm similar to clustering is used to calculate the distance from the test sample point to each type of center.
[0043] Furthermore, in the distance between the test sample point and each class center calculated by using a clustering-like nearest neighbor algorithm, the shortest class label is the label, and the center of the nth class in the cluster is calculated by the following formula:
[0044]
[0045] Using the properties of the kernel function, we can calculate its norm.
[0046]
[0047] Test sample φ(x) to the center φ of the nth class n Distance
[0048]
[0049] The decision function f(x) based on the clustering multi-core learning model and the test algorithm for model optimization satisfies the optimization formula:
[0050]
[0051] Furthermore, the stationary point method is used to optimize the solution in linear time to a simple linear optimization problem under the l1 norm (‖d‖1=1) constraint. The MATLAB optimization tool is used to obtain d m The optimal solution of .
[0052] A text emotion cause extraction system based on clustering multi-core learning includes a sampling module for generating samples containing text emotion causes based on text information recognition rules of clustering multi-core learning; a training module for generating a module of extracted emotion causes after learning, training and optimizing the emotion causes according to the recognition rules for the samples containing text emotion causes; and a testing module for testing the algorithm again using a clustering perspective for the data module optimized by the emotion cause training algorithm to obtain classification results and generate a data module for extracting emotion causes.
[0053] Furthermore, in the text emotion and reason extraction system based on clustering multi-core learning, the recognition rule is to form a data set and a test data set with training samples and test samples of sentences c stored in text form and containing emotions and reasons, and define the types of emotion and reason extraction p1, p2, ..., p n , using information extraction technology, after extracting the sentiment word w1 in sentence c, extract the sentiment clause c1 and the cause clause c2, locate the keyword w2 in the cause clause c2, and combine the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2. A clustering-based multi-core learning model is used to determine whether the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 constitute a given sentiment-cause relationship.
[0054] Beneficial Effects
[0055] Compared with the existing three types of emotion cause extraction methods, the advantages of the present invention are:
[0056] 1. Observe and solve classification problems from the perspective of clustering. This can combine the advantages of the two to derive a unified model. Given an unlabeled data set, S = {x1, x2, ..., x l} Find a mapping
[0057] f:S→{1, 2,…, N}.
[0058] The evaluation of clustering quality depends on two factors: first, the sample points of the same category should be close to each other, and the distance within the category should be as small as possible; second, the sample points of different categories should be separated enough, and the distance between categories should be as large as possible. A good model should take both into account.
[0059] First, a good clustering model should satisfy
[0060]
[0061] Among them, φ(x) is the projection function on the feature space F, which contains the algebraic sum of two parts. The former is the intra-class distance, and the latter is the inter-class distance. λ is the trade-off value. Formula (3) is derived,
[0062]
[0063] Where A is a constant when the data set is given. Therefore, substituting the above formula into (4), it can be expressed as
[0064]
[0065] Therefore, another measure of the clustering model is obtained.
[0066] Secondly, a good clustering model should satisfy
[0067]
[0068] This formula (6) only includes the calculation of intra-class distance.
[0069] As mentioned above, there are two main differences between classification and clustering problems, besides whether they are supervised or not:
[0070] First, the categories of classification problems are determined in advance, while the categories of clustering problems are not determined in advance.
[0071] Second, the classification problem does not consider whether there is cohesion within the same category, while the clustering problem needs to consider both the intra-class distance and the inter-class distance. If the number of categories is given, the first difference between the classification and clustering problems is eliminated. The second difference can be seen from formula (6), when the intra-class distance is small enough, the inter-class distance is also large enough.
[0072] 2. Its optimization objective function has a continuous lower bound on the domain of definition.
[0073] 3. It has been theoretically proven that the clustering-based multi-core learning model solves a linear programming problem under the l1 norm constraint. It does not require repeated iterations and can be solved in linear time.
[0074] 4. The clustering perspective makes it naturally suitable for multi-classification models. BRIEF DESCRIPTION OF THE DRAWINGS
[0075] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.
[0076] Figure 1 A flowchart of a text sentiment reason extraction method based on clustering multi-kernel learning
[0077] Figure 2 A hardware structure diagram of a text emotion extraction system based on clustering multi-core learning DETAILED DESCRIPTION
[0078] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments of the specification. The embodiments of the present invention are described below by specific specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.
[0079] It should be noted that the diagrams provided in the present embodiment are only used to illustrate the basic concept of the present invention in a schematic manner, so the diagrams only show the components related to the present invention rather than being drawn according to the number, shape and size of the components in actual implementation. The type, quantity and proportion of each component in actual implementation can be changed at will, and the component layout type may also be more complicated. The structure, proportion, size, etc. shown in the drawings attached to this specification are only used to match the content disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions that the present invention can be implemented, so they have no technical substantive significance. Any modification of the structure, change of the proportional relationship or adjustment of the size should still fall within the scope of the technical content disclosed by the present invention without affecting the effect and purpose that can be achieved by the present invention.
[0080] See also Figure 1 This embodiment provides a text emotion reason extraction method and system based on clustering multi-core learning, which is applied to the school mental health big data platform. Through the psychological growth and precision support system, through regular evaluation, the dynamic mental health status of each student and the comprehensive mental health status of all students are formed, and personal mental health electronic files are formed to provide personalized and intelligent solutions for management departments. The big data platform includes a psychological growth and precision support system (general testing platform), a mental health dynamic early warning platform, a mental health electronic file, a localized mental health online service platform (APP / applet), a mental health hotline operation, and offline mental health consultation. It integrates and calculates multivariate data to meet the needs of the psychological center, balance supply and demand, and realize the informatization of mental health work. The specific implementation steps are as follows:
[0081] S1. The psychological text data collected by the big data platform is stored in the form of documents as a data set, and each sample in the data set is a sentence c containing emotions and reasons;
[0082] S2, define several types of sentiment reason extraction p1, p2, ..., p n ;
[0083] S3, using information extraction technology, extract the sentiment word w1 in sentence c, and extract the sentiment sentence c1 and the cause clause c2 respectively;
[0084] S4, locate the keyword w2 in the cause clause c2;
[0085] S5, after combining the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2, use the relation extraction technology to perform multi-classification processing to determine whether it meets the given sentiment cause relationship p i ;
[0086] S6, the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 meet the given sentiment cause relationship p i , the reason clause c2 is the emotional reason clause to be extracted.
[0087] The main idea of kernel methods is based on the following assumption: "A point set that cannot be linearly separated in a low-dimensional space may become linearly separable when converted into a point set in a high-dimensional space." Its essence is to define a feature transformation (the inner product of the sample in the feature space) to map the sample set in the input space (non-linearly separable) to the reproducing kernel Hilbert space (inner product space), and then become a sample set in the feature space (linearly separable). The key to kernel methods is to find a suitable kernel function (base kernel).
[0088] Multi-kernel learning refers to the use of a set of predefined base kernels to form new kernels for learning through linear or nonlinear combination. It is suitable for situations where sample features contain heterogeneous information, the sample size is large, the multidimensional data is irregular, or the data is non-flatly distributed in the high-dimensional feature space. In the combined feature space constructed by multiple feature spaces, multi-kernel learning utilizes the feature mapping capabilities of each base kernel to more accurately express the data in the combined feature space. This solves the problem of insufficient base kernel representation ability and can significantly improve learning performance. The solution to multi-kernel learning problems often has to be transformed into complex forms such as semidefinite programming or quadratic constrained quadratic programming to solve.
[0089] Entity Relation Extraction (RE), also known as relation extraction, is an important research topic in the field of information extraction. The purpose of relation extraction is to extract pre-defined binary relations from unstructured natural language texts based on named entity extraction for further analysis. In recent years, relation extraction has become an important research direction that is indispensable for natural language understanding. The structured data obtained by relation extraction can be used in many fields such as automatic question-answering systems, semantic retrieval, ontology learning, semantic web annotation, etc.
[0090] Considering that the extraction of sentiment reasons can be converted into entity relationship extraction, it can essentially be regarded as a multi-classification problem. Therefore, the kernel method can be used and combined with clustering characteristics to solve it. This patent proposes for the first time a multi-kernel learning model based on clustering for sentiment reason extraction. Classification and clustering are relatively similar tasks in machine learning tasks. Both are to find a mapping f∶X→Y from training samples to categories. The classification problem is a supervised learning problem, and the number of categories is given in advance. The clustering problem is an unsupervised learning problem, and the number of categories cannot generally be determined in advance. The present invention observes and processes classification problems from a clustering perspective for the first time. After mapping the samples to a high-dimensional kernel space, the objective function of clustering is used as the objective function of classification, thereby constructing a multi-kernel learning model based on clustering.
[0091] The data collected by the mental health big data platform is stored in the form of documents as data sets, including training data sets and test data sets. The model is optimized based on the clustering multi-core learning model with a training algorithm. The multi-core learning model contains a given label data set.
[0092]
[0093] in represents the i-th training example of the n-th class, l n is the number of training examples of the nth class,
[0094] Y={1,2,…,N}
[0095] Multi-core combination function, where k m is the base kernel function, d m is the weight, φ(x) is the characteristic function of k, and the calculation
[0096]
[0097]
[0098] The clustering-based multi-core learning model is formalized as the following constrained optimization problem:
[0099]
[0100] Formula (1) is the model optimization proposed in this patent based on the clustering multi-kernel learning model and the training algorithm. m Given the situation, the stationary point method can be used to optimize the solution in linear time. It can be optimized in linear time as a simple linear optimization problem under the l1 norm (‖d‖1=1) constraint. In the simple l1 norm (‖d‖1=1) constraint linear optimization problem, a clustering-like nearest neighbor algorithm is used to calculate the distance from the test sample point to each center. In the use of a clustering-like nearest neighbor algorithm to calculate the distance from the test sample point to each center, the shortest class label is the label, and the center of the nth class in the cluster is calculated by the following formula
[0101]
[0102] Using the properties of the kernel function, we can calculate its norm.
[0103]
[0104] Test sample φ(x) to the center φ of the nth class n Distance
[0105]
[0106] The decision function f(x) for model optimization based on the clustering multi-kernel learning model and the test algorithm satisfies the following optimization formula, which is the test algorithm proposed by Benli.
[0107]
[0108] In this embodiment, the linear optimization problem under the constraint of simple l1 norm (‖d‖1=1) is solved by the stationary point method in linear time, and MATLAB optimization tool is used to obtain d m The optimal solution of .
[0109] like Figure 2 As shown, the present invention also provides a text emotion cause extraction system based on clustering multi-core learning, including a sampling module, which is used to generate samples containing text emotion causes based on text information recognition rules of clustering multi-core learning; a training module, which is used to optimize the emotion causes according to the recognition rules for the samples containing text emotion causes, and then generate a data module of extracted emotion causes; a testing module, which is used to optimize the data module after the emotion cause training algorithm is optimized using a clustering perspective, and then generate a module of extracted emotion causes.
[0110] Here, the recognition rule is to form a data set and a test data set for the training examples and test examples of sentences c stored in text form and containing emotions and reasons, and define the types of emotion reason extraction p1, p2, ..., p n ,After extracting the sentiment word w1 from sentence c by using information extraction technology, the sentiment clause c1 and the cause clause c2 are extracted, the keyword w2 in the cause clause c2 is located, the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 are combined, and the clustering-based multi-core learning model is used to determine whether the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 constitute a given sentiment-cause relationship.
[0111] By using the text emotion cause extraction method and system based on clustering multi-kernel learning to process the text of the mental health big data platform, it is possible to clearly and accurately extract the emotion cause data of each student participating in the data collection of the mental health big data platform. Through precise evaluation and regular assessment, the dynamic mental health status of each student and the comprehensive mental health status of all students are formed. At the same time, a personal mental health electronic file is formed, which realizes more accurate text emotion cause extraction, makes accurate early warning analysis for psychological crises, and has high industrial utilization value.
[0112] In summary, the present invention provides a method and system for extracting text emotion causes based on clustering multi-core learning, which takes into account the formalization problem that the sample points of the same category are close, the distance within the category is as small as possible, and the sample points of different categories are sufficiently separated, and the distance between categories is as large as possible. Its optimization objective function is continuous and has a lower bound on the definition domain, and it has been theoretically proved that the clustering-based multi-core learning model solves a linear programming problem under the l1 norm constraint, it does not require repeated iterations, and can be solved in linear time. It effectively overcomes the various shortcomings of the emotion cause extraction method for a given document in the field of psychological mental health assessment in the prior art, realizes more accurate text emotion cause extraction, makes accurate early warning analysis for psychological crises, and has high industrial utilization value.
Claims
1. A text sentiment reason extraction method based on clustering multi-kernel learning, based on clustering perspective observation and processing classification problems, after mapping samples to high-dimensional kernel space, the clustering objective function is used as the classification objective function to construct a clustering multi-kernel learning model, characterized in that: The emotional reason extraction method based on clustering multi-kernel learning model includes the following steps: S1. Storing psychological text data in the form of documents as a data set, where each sample in the data set is a sentence c containing emotions and reasons; S2. Define several types of emotion reason extraction p1, p2, ..., p n ; S3, using information extraction technology, extract the sentiment word w1 in sentence c, and extract the sentiment sentence c1 and the cause clause c2 respectively; S4, locate the keyword w2 in the cause clause c2; S5, after combining the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2, multi-classification processing is performed using relation extraction technology to determine whether the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 meet the given sentiment cause relationship p. i ; S6, the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 meet the given sentiment cause relationship p i , the reason clause c2 is the emotional reason clause to be extracted.
2. According to the method for extracting text emotion causes based on clustering multi-core learning according to claim 1, it is characterized in that: The data set includes a training data set and a test data set.
3. According to claim 2, a text emotion cause extraction method based on clustering multi-core learning is characterized in that: The clustering multi-core learning model is optimized by a training algorithm, and the clustering multi-core learning model includes a given label data set. in represents the i-th training example of the n-th class, l n is the number of training examples of the nth class, Y = {1, 2, ..., N} Multi-core combination function, where k m is the base kernel function, d m is the weight, φ(x) is the characteristic function of k, and the calculation The clustering-based multi-core learning model is formalized as the following constrained optimization problem: 。 4. According to claim 3, a text emotion cause extraction method based on clustering multi-core learning is characterized in that: The clustering multi-kernel learning model is optimized by the test algorithm, and the base kernel function k m Given, the solution is optimized in linear time by the stationary point method.
5. According to the method for extracting text emotion causes based on clustering multi-core learning according to claim 4, it is characterized in that: The base kernel function k m Given, the stationary point method is used to optimize and solve the problem in linear time as a simple linear optimization problem under the l1 norm (||d||1=1) constraint.
6. According to the method for extracting text emotion causes based on clustering multi-core learning according to claim 5, it is characterized in that: The simple linear optimization problem under the l1 norm (||d||1=1) constraint adopts a clustering-like nearest neighbor algorithm to calculate the distance from the test sample point to each type of center.
7. According to claim 6, a text emotion cause extraction method based on clustering multi-core learning is characterized in that: The class with the shortest distance is labeled, and the center of the nth class in the cluster is calculated by the following formula Using the properties of the kernel function, we can calculate its norm. Test example φ(x) to the center φ of the nth class n Distance The decision function f(x) based on the clustering multi-core learning model and the test algorithm for model optimization satisfies the optimization formula:
8. According to claim 5, a text emotion cause extraction method based on clustering multi-core learning is characterized in that: The stationary point method is used to optimize and solve the linear optimization problem in linear time as a simple l1 norm (||d||1=1) constraint. The optimization tool of MATLAB is used to obtain d m The optimal solution of .
9. A text emotion cause extraction system based on clustering multi-kernel learning, characterized in that: It includes a sampling module for generating samples containing text sentiment reasons based on text information recognition rules based on clustering multi-kernel learning; The training module is used to optimize the training algorithm for the emotional reasons according to the recognition rules for the samples containing the text emotional reasons, and then generate a data module of the extracted emotional reasons; the testing module is used to test the algorithm again from a clustering perspective on the data module after the emotional reason training algorithm is optimized, obtain the classification results, and generate a data module for extracting the emotional reasons.
10. A text emotion cause extraction system based on clustering multi-core learning according to claim 9, characterized in that: The recognition rule is to form a data set and a test data set with training samples and test samples of sentences c stored in text form and containing emotions and reasons, and define the types of emotion reason extraction p1, p2, ..., p n ,After extracting the sentiment word w1 from sentence c by using information extraction technology, the sentiment clause c1 and the cause clause c2 are extracted, the keyword w2 in the cause clause c2 is located, the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 are combined, and the clustering-based multi-core learning model is used to determine whether the sentiment word w1 in sentence c and the keyword w2 in the cause clause c2 constitute a given sentiment-cause relationship.