A multi-topic word vector text clustering method, system, terminal and storage medium
By calculating the mean word vectors and cosine distance between semantic center words and related words, a specified direction clustering of multi-topic text clustering is achieved, which solves the problems of high randomness in topic selection and single clustering topic in existing technologies, and improves clustering efficiency and accuracy.
Patent Information
- Application Number
- CN202310559795.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-15
- Publication Date
- 2026-01-02
- Estimated Expiration
- 2043-05-15
AI Technical Summary
Existing text clustering models cannot achieve multi-topic text clustering for a specified classification topic, and manual annotation is costly and inefficient, failing to meet the needs of multi-dimensional and multi-level topic-oriented clustering.
By calculating the average word vectors of the semantic center word and related words, the average word vector of the target topic is obtained, and the semantic information that matches the target topic is filtered out using the cosine distance between the word vectors, and text clustering is performed.
It enables multi-topic text clustering based on specified topics without the need for pre-labeling and model training, improving the accuracy and efficiency of clustering and solving the problems of high randomness in topic selection and single clustering topics in traditional models.
Smart Images
Figure CN116821330B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of text classification methods, and in particular to a multi-topic word vector text clustering method, system, terminal and storage medium. BACKGROUND
[0002] In actual production or scientific research, when text classification processing is involved, text clustering processing is often needed. Usually, the user needs to cluster the text according to certain specific topics. Moreover, since the text may contain multiple dimensions of topics, the same topic may contain multiple levels of sub-topics, so a method is needed to achieve this.
[0003] The existing mainstream text clustering topic model cannot achieve this purpose. The most common LDA topic model and K-Means classification model can only automatically cluster according to the predicted classification category number parameter. As for the specific topics of the clustered topics, the algorithm randomly generates them according to the text situation, and cannot realize topic-oriented text classification. For multi-topic modeling, although the Multi-Grain Topic model can preserve multi-dimensional and multi-level topics of clustered topics, it still cannot solve the pain point of specifying classification topics. Although the word embedding model represented by the word vector model can improve the classification accuracy when combined with the topic model, for example, the Embedded Topic Model, it still cannot solve the problem of specifying the classification direction. Deep learning classification models can partially achieve multi-topic text clustering of specified topics through supervised learning with massive training corpus and artificial classification labels, but the cost of artificial labeling is very high, and the efficiency is low when the specified topic is changed and artificial labeling is needed again.
[0004] Therefore, a multi-topic word vector text clustering method that can be topic-oriented is needed. SUMMARY
[0005] In order to cluster text containing multi-topic words, the present application provides a multi-topic word vector text clustering method, system, terminal and storage medium.
[0006] In the first aspect, the multi-topic word vector text clustering method provided by the present application adopts the following technical solution:
[0007] A multi-topic word vector text clustering method comprises:
[0008] performing word segmentation on the target text to obtain word segmentation information and obtaining word vector information corresponding to each word information;
[0009] obtaining semantic central words of the target text;
[0010] calculating the word vector mean value between the semantic central word and the associated word to obtain a semantic word vector;
[0011] obtaining a plurality of target words related to the target theme;
[0012] calculating the word vector mean value between the target words as a target theme average word vector;
[0013] calculating the word vector cosine distance information between the target theme average word vector and the semantic word vector in each text, and retaining the word information corresponding to the larger word vector cosine distance information through a preset proportional relationship as the semantic theme information conforming to the target theme;
[0014] performing text clustering on the semantic theme information conforming to the target theme to obtain a clustering result.
[0015] First, in order to extract the semantic theme conforming to the specified theme, a series of semantic themes need to be generated for the text, and all possible semantic themes are extracted as much as possible. At the same time, the specified direction of text clustering needs to be specified. Thereafter, the semantic themes conforming to the target theme are screened by calculating the cosine distance information between the target theme average word vector and the semantic word vector in each text. Finally, the semantic themes obtained are classified again through text clustering, and the required multi-theme word text clustering is completed. Thus, the user can cluster the text into several categories according to one or more specified themes without pre-labeling and model training.
[0016] Preferably, the method for obtaining the semantic central word of the target text comprises:
[0017] According to the obtained word vector information, the feature vector center degree of each word information is calculated;
[0018] By comparing the feature vector semantic center degree of each word information, a plurality of word information is selected as the semantic central word.
[0019] Preferably, the method for calculating the feature vector center degree of each word information comprises:
[0020] The word vector cosine distance information between the word vector of each word information and the word vector of another word information is calculated;
[0021] The word information corresponding to the larger part of all word vector cosine distance information is retained through a preset proportional relationship;
[0022] The feature vector semantic center degree of the retained word information is calculated according to the word vector cosine distance information.
[0023] By adopting the above technical solutions, the core idea of the feature vector centrality is adopted to construct the feature vector centrality, and the words at the semantic theme centers are obtained by calculating the feature vector semantic centrality of each word. In this way, the most semantic theme representative words, called semantic center words, can be extracted. Each semantic center word will also represent a semantic theme.
[0024] Preferably, the method for calculating the word vector cosine distance information follows:
[0025]
[0026] Strength f,z = CosineSim(F, Z)
[0027] Wherein, the F and Z are two word vector information respectively, CosineSim is a cosine distance calculation function, Strength f,z represents the word vector cosine distance of the word f and the word z.
[0028] Preferably, the method for calculating the feature vector semantic centrality of the reserved word information according to the word vector cosine distance information includes:
[0029]
[0030] Wherein, Central i,f is the feature vector semantic centrality of the word f in the i document, wherein cw represents the network association with the word f, n is the number of associated words, and CW represents the total number of words f.
[0031] Preferably, the method for calculating the word vector mean between the target words as the target theme average word vector includes:
[0032]
[0033] Wherein, is the target theme average word vector, t is the number of target words obtained in relation to the target theme, [x1, x2, …, x 200 ] is a 200-dimensional word vector.
[0034] By adopting the above technical solutions, due to the characteristics of the word vector model, the mean of several theme-related word vectors can better represent the semantic theme than the single word vector.
[0035] Preferably, whether the required theme information is in the clustering result is judged;
[0036] If not, supplementary target vocabulary related to the required theme information is obtained in the target text, and the word vector mean between the theme information and the supplementary target vocabulary is calculated as a reserved theme average word vector;
[0037] The vocabulary information corresponding to the reserved theme average word vector is added to the semantic theme information, and text clustering is performed again to obtain a clustering result.
[0038] By adopting the above technical solution, the required semantic theme can be forcibly reserved, so that the required semantic theme does not disappear after classification is completed.
[0039] In a second aspect, the multi-theme word vector text clustering system provided by the present application adopts the following technical solution:
[0040] A multi-theme word vector text clustering system comprises,
[0041] A text preprocessing module is configured to perform word segmentation on a target text to obtain vocabulary information after word segmentation, and obtain word vector information corresponding to each vocabulary information.
[0042] A target text semantic theme extraction module is configured to obtain semantic center vocabulary of the target text, and calculate the word vector mean between the semantic center vocabulary and the associated vocabulary to obtain a semantic word vector.
[0043] A target theme extraction module is configured to obtain a plurality of target vocabulary related to a target theme,
[0044] The word vector mean between the target vocabulary is calculated as a target theme average word vector,
[0045] The word vector cosine distance information between the target theme average word vector and the semantic word vector in each text is calculated, and the vocabulary information corresponding to the larger word vector cosine distance information is reserved through a preset proportional relationship as the semantic theme information conforming to the target theme.
[0046] A text clustering module is configured to perform text clustering on the semantic theme information conforming to the target theme to obtain a clustering result.
[0047] In a third aspect, the intelligent terminal provided by the present application adopts the following technical solution:
[0048] An intelligent terminal comprises a memory and a processor, and the memory stores a computer program capable of being loaded and executed by the processor to perform the above multi-theme word vector text clustering method.
[0049] In a fourth aspect, the computer storage medium provided by the present application can store a corresponding program, and adopts the following technical solution:
[0050] A computer readable storage medium stores a computer program capable of being loaded by a processor and executing any one of the above multi-topic word vector text clustering methods.
[0051] In summary, the application firstly aggregates text content into a hierarchy of semantic topics, defines all possible semantic topics, represents the text with qualified semantic topics, and matches with a preset target topic to obtain text clustering according to the specified topic classification. Thus, the key problems of traditional topic classification, such as randomness in topic selection, inability to specify topics as needed, and single clustering topic, are improved. BRIEF DESCRIPTION OF DRAWINGS
[0052] Figure 1 is a flowchart of a multi-topic word vector text clustering method of one embodiment of the application.
[0053] Figure 2 is a specific calculation method of the feature vector centrality of one embodiment of the application. DETAILED DESCRIPTION
[0054] The following will be described in detail below with reference to the accompanying drawings Figure 1 and 2 The application will be further described in detail.
[0055] The embodiment of the application discloses a multi-topic word vector text clustering method, which comprises the following steps: performing word segmentation on target text to obtain word segmentation information and obtaining word vector information corresponding to each word information; calculating the feature vector centrality of each word information according to the obtained word vector information; selecting a plurality of word information as semantic center words by comparing the feature vector semantic centrality of each word information; calculating the word vector mean value between the semantic center words and the associated words to obtain semantic word vectors; obtaining a plurality of target words related to a target topic; calculating the word vector mean value between the target words as a target topic average word vector; calculating the word vector cosine distance information between the target topic average word vector and the semantic word vectors in each text, and retaining the word information corresponding to the larger word vector cosine distance information through a preset proportional relationship, as the semantic topic information meeting the target topic; and performing text clustering on the semantic topic information meeting the target topic to obtain a clustering result.
[0056] The embodiment of the application aims to realize a text clustering method that can retain multiple dimensional classifications for a series of text data with a pre-specified topic. The user can cluster the text into several categories according to one or more specified topics without pre-labeling and model training. It should be noted that the classification result can be multiple text categories under a single topic, or multiple topics each containing several text categories.
[0057] For example, a user needs to cluster several hundred thousand news texts according to two categories of "wine" and "sales and management" to identify which texts meet the topics, and further cluster the texts into subcategories. For the text: "Under the unremitting efforts of the distributors, the sales of Maotai and Wuliangye did not decrease significantly, while the prices and sales of low-end white wine such as Jiang Xiaobai and Erguotou increased significantly. All this is due to the appropriate online publicity and the smooth offline sales channels. However, the sales of red wine have temporarily stagnated." If the categories are preset as "wine" and "sales and management", the algorithm will classify this text into the categories of "wine", subcategory "white wine" and "red wine", and "sales and management", subcategory "sales channel", etc. Thus, the key problems of traditional topic classification, such as randomness in topic selection, inability to specify topics as needed, and single clustering topic, are solved.
[0058] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.
[0059] In addition, the term "and / or" in this paper is only to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents an "or" relationship between the associated objects unless otherwise specified.
[0060] The embodiments of the present application will be described in further detail below with reference to the drawings of the specification.
[0061] Referring to Figure 1 The embodiments of the present application provide a multi-topic word vector text clustering method, and the main process of the method is described as follows.
[0062] Step S100: cutting the target text to obtain word information after cutting, and obtaining word vector information corresponding to each word information.
[0063] The target text generally refers to a single document, which can be obtained from a webpage or other public information source, or can be manually collected and input. The first step of word segmentation is to remove irrelevant information such as punctuation marks that may affect word segmentation, and then use the "full mode" of the pythonJieba segmentation library and a pre-trained dictionary to force the preservation of word groups to preserve the text semantics as much as possible. The pre-trained dictionary can be a dictionary pre-built according to the needs, or the vocabulary table of the "Tencent AI-LAB word vector model".
[0064] For example, "financial technology" will be segmented into "finance" and "technology" in the ordinary word segmentation mode. However, using the pre-trained dictionary and the "full mode" method, "financial technology" will be segmented into "finance", "technology", and "financial technology", thus preserving the unique semantics of the word group "financial technology".
[0065] After word segmentation, each word vector is assigned a word vector from the word vector model. The specific dimension of the word vector can be selected according to actual needs. In this embodiment, a 200-dimensional word vector is used by default to describe the relevant steps. The word vector model can be trained by itself, or the open source "Tencent AI-LAB word vector model" can be used.
[0066] Step S200: Obtain the semantic center word of the target text.
[0067] The semantic center word represents the representative word that best matches the semantic theme of the text, that is, each semantic center word also represents a semantic theme. Specifically, the characteristic vector center degree is used to obtain the required semantic center word. According to the obtained word vector information, the characteristic vector center degree of each word information is calculated, and several word information is selected as the semantic center word by comparing the characteristic vector semantic center degrees of the word information. For the specific calculation method of the characteristic vector center degree, refer to Figure 2 , including:
[0068] Step S210: Calculate the word vector of each word information and the word vector of another word information.
[0069] Step S211: Retain the word information corresponding to the larger part of all word vector cosine distance information through a predetermined proportional relationship.
[0070] In the specific feature vector center degree calculation method, each word in the text is taken as a network node, the cosine distance between word vectors is taken as the association and association strength between network nodes, and only the word information meeting the requirements is retained to represent the relationship between words. As a specific calculation method, it meets:
[0071]
[0072] Strength f,z = CosineSim(F, Z)
[0073] In the formula, F and Z are two word vector information respectively, CosineSim is the cosine distance calculation function, and Strength f,z represents the cosine distance between word vectors of word f and word z.
[0074] Generally, only the top 10% cosine distance is retained to represent the association between words in actual operation. The 10% cosine distance indicates that the association between words is strong, and the stronger the association degree, the more important the word plays in the document. The 10% proportion relationship can be changed according to actual needs, such as 5% or 15%.
[0075] Step S212: Calculate the feature vector semantic center degree of the retained word information according to the word vector cosine distance information.
[0076] In which, the specific calculation steps are as follows:
[0077]
[0078] In the formula, Central i,f is the feature vector semantic center degree of word f in i document, in which cw represents the network association with word f, n is the number of associated words, CW is the total number of word f, and the number of associated words is determined by the retained word vector cosine distance. Each retained word vector cosine distance corresponds to two words representing the association.
[0079] According to the formula, the feature vector semantic center degree of the word is calculated by calculating the average distance between the word and the nodes around the network node.
[0080] For example, the word "Maotai" as a network node in the semantic network, its associated network nodes are "Wuliangye", "Er Guo Tou" and other nodes, and there is a very high correlation between these nodes, so the semantic centrality of the feature vector of "Maotai" will also be higher. After calculating the semantic centrality of each feature vector, the word information with higher semantic centrality can be selected as the semantic center word by corresponding screening method, for example, several word information with higher semantic centrality can be selected as the semantic center word according to the proportion.
[0081] Step S300: Calculate the word vector mean between the semantic center word and the associated word to obtain the semantic word vector.
[0082] Among them, due to the characteristics of the word vector model, the mean of several topic-related word vectors can better represent the semantic topic than the single word vector. The word vector mean here refers to the average of the word vectors of each word used for calculation, and the obtained word vector is the required semantic word vector.
[0083] For example, the word vector of the word "Maotai" will be averaged with the word vectors of the words "Wuliangye", "Er Guo Tou" and other words to obtain the "semantic word vector" of "Maotai". Thus, after this step, each semantic topic (i.e. semantic center word) and semantic word vector in the text can be obtained. By analogy to all the texts to be processed, the semantic topic and semantic word vector of each text can be obtained.
[0084] Step S400: Obtain several target words related to the target topic.
[0085] Among them, in order to perform text clustering in a specified direction, the clustering topic direction needs to be specified. Due to the characteristics of the word vector model, the mean of several topic-related word vectors can better represent the semantic of the target topic than the single word vector. For example, the word vector of the word "wine" may be highly close to the word vectors of "alcohol" and "drunk driving", which are not target words. However, the average word vector of the words "baijiu", "red wine" and "beer" can better represent the semantic of the "wine" category represented by the word "wine". Therefore, for the setting of a certain target topic, several target words related to the target topic need to be given. The target topic is determined by a human pre-given manner.
[0086] Step S500: Calculate the word vector mean between the target words as the target topic average word vector.
[0087] Specifically, it satisfies:
[0088]
[0089] Among them, is the target topic average word vector, t is the number of target words related to the target topic, [x1, x2, …, x 200 ] is a 200-dimensional word vector.
[0090] Step S600: Calculate the word vector cosine distance information between the target topic average word vector and the semantic word vector in each text, and retain the word vector information corresponding to the larger word vector cosine distance information through a preset proportional relationship, as the semantic topic information conforming to the target topic.
[0091] Wherein, the calculation method of the word vector cosine distance information between the semantic word vectors in this step is the same as step S211, and similarly, for the preset proportional relationship to retain the word vector information corresponding to the larger word vector cosine distance information, only the first 10% of the cosine distance can be retained to represent the target topic average word vector and the semantic word vector in each text has a larger correlation, or 5% or 15% proportion. After that, the combination of the retained word vectors contains the target topic average word vector, that is, the required target topic average word vector, and the word information corresponding to these target topic word vectors is the semantic topic information conforming to the target topic.
[0092] Step S700: Text clustering is performed on the semantic topic information conforming to the target topic to obtain a clustering result.
[0093] Wherein, the text clustering method here can be K-Means clustering, or common text clustering methods such as L-DA. Taking K-Means clustering as an example, after K-Means clustering of the semantic topic information conforming to the target topic, the preset K categories of semantic topics conforming to the target topic can be obtained. Further, each text can be traversed and filtered for semantic topics conforming to the target topic before clustering. If the text contains a certain target topic conforming to the semantic topic, the text is defined as a certain topic, and it is finally clustered.
[0094] At this time, after the text is clustered in step S700, the clustering result cannot identify the required target topic, and a specific semantic topic can be further forced to retain and cluster. Specifically, it includes:
[0095] Step S800: Determine whether the clustering result contains the required topic information.
[0096] Wherein, the judgment here can be artificial judgment, or through machine selection to judge whether the result contains the pre-set required topic information.
[0097] Step S810: If not, obtaining supplementary target vocabulary related to the required topic information in the target text, and calculating the word vector mean value between the topic information and the supplementary target vocabulary as the reserved topic average word vector.
[0098] The execution mode of step S810 is the same as that of step S500, except that the processed word vector is changed from the target topic-related several target vocabulary to the topic information-related supplementary target vocabulary.
[0099] Step S820: Adding the vocabulary information corresponding to the reserved topic average word vector to the semantic topic information and re-performing text clustering to obtain a clustering result.
[0100] In this step, the average word vector of the required semantic topic to be forcibly reserved is obtained, and the semantic topic cosine distance screening judgment process is directly performed together with the rest of the semantic topic information. This way can reserve the required semantic topic, so that the result meets the required requirements.
[0101] Based on the same inventive concept, the embodiments of the present application also disclose a multi-topic word vector text clustering system, which comprises:
[0102] A text preprocessing module is configured to perform word segmentation on a target text to obtain segmented vocabulary information, and obtain word vector information corresponding to each vocabulary information;
[0103] A target text semantic topic extraction module is configured to obtain semantic center vocabulary of the target text, and calculate the word vector mean value between the semantic center vocabulary and the associated vocabulary to obtain a semantic word vector;
[0104] A target topic extraction module is configured to obtain several target vocabulary related to a target topic,
[0105] Calculate the word vector mean value between the target vocabulary as the target topic average word vector,
[0106] Calculate the word vector cosine distance information between the target topic average word vector and the semantic word vector in each text, and retain the vocabulary information corresponding to the larger word vector cosine distance information through a preset proportional relationship, as the semantic topic information conforming to the target topic;
[0107] A text clustering module is configured to perform text clustering on the semantic topic information conforming to the target topic to obtain a clustering result.
[0108] Based on the same inventive concept, the embodiments of the present application provide an intelligent terminal, comprising a memory and a processor, the memory stores a computer program capable of being loaded and executed by the processor, and the computer program performs the multi-topic word vector text clustering method as shown in the figure.
[0109] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional modules is taken as an example, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.
[0110] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules or units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.
[0111] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.
[0112] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.
[0113] The embodiment of the present application also provides a computer readable storage medium, which stores a program capable of being loaded and executed by a processor to realize each step described in the flowchart.
[0114] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the entire or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.
[0115] The above embodiments are only used to introduce the technical solutions of the present application in detail, but the above embodiment descriptions are only used to help understand the method of the present application and its core idea, and should not be understood as a limitation of the present application. Those skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.
Claims
1. A multi-topic word vector text clustering method, characterized in that, The method comprises the following steps: performing word segmentation on the target text to obtain word information after word segmentation, and obtaining word vector information corresponding to each word information; obtaining semantic center words of the target text; calculating the word vector mean value between the semantic center words and related words to obtain semantic word vectors; obtaining a plurality of target words related to the target theme; calculating the word vector mean value between the target words as the target theme average word vector; calculating the word vector cosine distance information between the target theme average word vector and the semantic word vector in each text, and retaining the word information corresponding to the larger word vector cosine distance information through a preset proportional relationship, as the semantic theme information conforming to the target theme; performing text clustering on the semantic theme information conforming to the target theme to obtain a clustering result; The method for obtaining the semantic center words of the target text comprises the following steps: calculating the feature vector center degree of each word information according to the obtained word vector information; and selecting a plurality of word information as semantic center words by comparing the feature vector semantic center degrees of the word information; The method for calculating the feature vector center degree of each word information comprises the following steps: calculating the word vector cosine distance information between the word vector of each word information and the word vector of another word information; retaining the word information corresponding to the larger part of all word vector cosine distance information through a preset proportional relationship; and calculating the feature vector semantic center degree of the retained word information according to the word vector cosine distance information; The method for calculating the word vector cosine distance information comprises the following steps: Strength f,z = CosineSim(F, Z) Wherein, the F and Z are two word vector information respectively, CosineSim is a cosine distance calculation function, Strength f,z The cosine distance of the word vector of the representative word f and the word z; The method for calculating the feature vector semantic center degree of the retained word information according to the word vector cosine distance information comprises the following steps: where Central i,f is the semantic centrality of the feature vector of a term f in the i document, where cw represents the number of network connections with the term f, n is the number of associated terms, and CW is the total number of terms f.
2. The multi-thematic term vector text clustering method according to claim 1, characterized in that, The method for calculating the word vector mean value between the target words as the target theme average word vector comprises the following steps: wherein, is the average word vector of the target topic, t is the number of target words related to the target topic, [x1, x2, …, x 200 ] is a 200-dimensional word vector.
3. The multi-theme word vector text clustering method according to claim 1, characterized in that, determining whether there is the required theme information in the clustering result; if not, obtaining supplementary target words related to the required theme information in the target text, and calculating the word vector mean value between the theme information and the supplementary target words as a reserved theme average word vector; adding the word information corresponding to the reserved theme average word vector to the semantic theme information and re-performing text clustering to obtain a clustering result.
4. A multi-topic word vector text clustering system, characterized by, The system is used for executing the multi-theme word vector text clustering method according to claim 1, comprising: a text preprocessing module configured to perform word segmentation on the target text to obtain word information after word segmentation, and obtain word vector information corresponding to each word information; a target text semantic theme extraction module configured to obtain semantic center words of the target text, and calculate the word vector mean value between the semantic center words and related words to obtain semantic word vectors; a target theme extraction module configured to obtain a plurality of target words related to the target theme, and calculate the word vector mean value between the target words as the target theme average word vector, calculate the word vector cosine distance information between the target theme average word vector and the semantic word vector in each text, and retain the word information corresponding to the larger word vector cosine distance information through a preset proportional relationship, as the semantic theme information conforming to the target theme; The text clustering module is configured to perform text clustering on the semantic topic information conforming to the target topic to obtain a clustering result.
5. A smart terminal, characterized in that, A computer program product comprising a memory and a processor, said memory having stored thereon a computer program loadable and executable by the processor to perform the method of any of claims 1 to 3.
6. A computer-readable storage medium, characterized in that, A computer program product comprising a memory and a processor, said memory having stored thereon a computer program loadable and executable by the processor to perform the method of any of claims 1 to 3. A computer program product comprising a memory and a processor, said memory having stored thereon a computer program loadable and executable by the processor to perform the method of any of claims 1 to 3.