A method and system for generating user topic tags based on WeChat Work discussion groups

Abstract

This application relates to a method and system for generating user topic tags based on WeChat Work discussion groups. The method includes: a word vector training step, where different discussion tags are preset according to industry type, and corresponding external word vectors are obtained by filtering them; the external word vectors are combined with conversation archives to train word vectors; a priority word acquisition step, where the similarity between the word vectors and the discussion tags is calculated, and word vectors that meet preset conditions are added to a priority word list; a conversation tagging step, where conversation archives are scanned using a word segmentation system and tagged according to priority words and preset stop words; and a topic tag generation step, where the tagging results are processed to obtain user topic tags. This application enables the rapid construction of user discussion topic tags from massive amounts of user conversation information.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for generating user topic tags based on enterprise WeChat discussion groups. Background Technology

[0002] In enterprise WeChat services targeting consumers (toC), company operations staff typically use WeChat Work to add customers to discussion groups and conduct marketing campaigns. In this context, operations staff can proactively market through these discussion groups by responding to user topics and promoting the efficacy of their products. Furthermore, users can also engage in discussions on everyday topics within these groups.

[0003] For text messages discussed by users in WeChat Work discussion groups, operations staff can use WeChat Work's conversation archiving function to record them. Correctly identifying the topics discussed by users can help operations staff build user tags and, based on these tags, conduct marketing campaigns for different user categories, or discover user discussion topics to assist in building operational materials.

[0004] Traditionally, building user tags based on discussion group information typically involves operations staff manually reviewing chat messages and assigning tags. However, this method has the following bottlenecks:

[0005] When the amount of data archived in WeChat Work conversations reaches a certain level (e.g., multiple data entries, multiple discussion groups, etc.), manual tagging becomes slow. At the same time, different operations personnel have different user tagging standards, and it is impossible to determine whether the topics discussed by users are popular topics. This may result in long-tail data being generated (e.g., a tag only matches one person), making it difficult to help operations personnel carry out subsequent marketing activities.

[0006] Currently, no effective solution has been proposed to address the slow progress of manual marking in related technologies. Summary of the Invention

[0007] This application provides a method and system for generating user topic tags based on enterprise WeChat discussion groups, in order to at least solve the problem of slow manual tagging in related technologies.

[0008] In a first aspect, embodiments of this application provide a method for generating user topic tags based on WeChat Work discussion groups, including the following steps:

[0009] The word vector training process involves pre-setting different discussion tags based on industry type, and then selecting corresponding external word vectors. These external word vectors are then combined with conversation archives to train the word vectors.

[0010] The priority word acquisition step involves calculating the similarity between word vectors and discussion tags, and adding word vectors that meet preset conditions to the priority word list;

[0011] The conversation tagging process involves scanning the conversation archive using a word segmentation system and tagging the conversation archive based on priority words and preset stop words.

[0012] The topic tag generation process involves processing the tagging results to obtain user topic tags.

[0013] In some embodiments, the word vector training step further includes:

[0014] The external word vector selection step involves obtaining Tencent AI Lab Ab word vectors from Tencent AI Lab, calculating the Euclidean distance between the Tencent AI Lab Ab word vectors and the corresponding vectors of the discussion tags, and selecting external word vectors accordingly.

[0015] The word vector output step involves preprocessing the conversation archive to obtain the corresponding one-hot vectors, inputting the one-hot vectors and external word vectors into the Word2vec model, and outputting word vectors.

[0016] In some embodiments, the priority word acquisition step specifically includes:

[0017] The cosine similarity between the word vector and the word vector corresponding to the discussion tag is calculated using the following formula.

[0018]

[0019] Where A and B represent word vectors and word vectors corresponding to discussion tags, respectively, n represents the total dimension, and i represents the i-th dimension.

[0020] When the cosine similarity is greater than a preset threshold, the word segment corresponding to the word vector is added to the priority word list.

[0021] In some embodiments, the session tagging step specifically includes:

[0022] The conversation archive scanning process involves adding priority words and preset stop words to the jieba word segmentation system and then using the jieba word segmentation system to scan the conversation archive.

[0023] The information output step involves deleting the corresponding word segments from the session archive based on preset stop words, and outputting the corresponding priority words, discussion tags, and speaking users that are matched in the remaining part of the session archive.

[0024] In some embodiments, the topic tag generation step specifically includes:

[0025] The information deduplication step deduplicates the corresponding priority words and speaking users output in the information output step, and then obtains the hit priority words and the number of corresponding speaking users.

[0026] The low-frequency data cleaning steps involve calculating the variance and mean of the number of hit priority words and the number of users who spoke, and then filtering using the 3*sigma principle.

[0027] The topic tag generation process involves outputting the discussion tag information of the speaking user based on the filtering results, and obtaining the user's topic tag accordingly.

[0028] In some embodiments, the filtering conditions in the low-frequency data cleaning step further include:

[0029] If the number of users who speak when a priority word is hit is less than a set value, then the priority word will be filtered out. The set value is the mean minus 3 * standard deviation.

[0030] Secondly, embodiments of this application provide a user topic tag generation system based on enterprise WeChat discussion groups, applying the user topic tag generation method of the first aspect, including:

[0031] The word vector training module presets different discussion tags according to industry type and selects corresponding external word vectors accordingly. The external word vectors are combined with the conversation archive to train word vectors.

[0032] The priority word acquisition module calculates the similarity between word vectors and discussion tags, and adds word vectors that meet preset conditions to the priority word list;

[0033] The conversation tagging module involves the following steps: based on priority words and preset stop words, the word segmentation system scans the conversation archive and tags the conversation archive.

[0034] The topic tag generation module processes the tagging results to obtain user topic tags.

[0035] In some embodiments, the topic tag generation module includes:

[0036] The information deduplication unit receives the corresponding priority words and speaking users output by the session tagging module and deduplicates them to obtain the hit priority words and the number of corresponding speaking users;

[0037] The low-frequency data cleaning unit calculates the variance and mean of the number of hit priority words and the number of speaking users, and then filters the data using the 3*sigma principle.

[0038] The topic tag generation unit outputs the discussion tag information of the speaking user based on the filtering results, and obtains the user's topic tag accordingly.

[0039] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the user topic tag generation method based on enterprise WeChat discussion groups as described in the first aspect above.

[0040] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the user topic tag generation method based on Enterprise WeChat discussion groups as described in the first aspect above.

[0041] Compared to related technologies, the user topic tag generation method and system based on WeChat Work discussion groups provided in this application can be applied to the fields of data processing technology and data mining technology. Through this patent, user discussion topic tags can be quickly constructed from massive amounts of user conversation information, which greatly helps operators to tag users quickly and accurately, and prepare for downstream marketing activities.

[0042] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0043] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0044] Figure 1 This is a flowchart of a user topic tag generation method based on Enterprise WeChat discussion groups according to an embodiment of this application;

[0045] Figure 2 This is a flowchart of the word vector training steps according to an embodiment of this application;

[0046] Figure 3 This is a flowchart of the session tagging steps according to an embodiment of this application;

[0047] Figure 4 This is a flowchart of the session tagging steps according to an embodiment of this application;

[0048] Figure 5 This is a flowchart of another method for generating user topic tags based on enterprise WeChat discussion groups according to an embodiment of this application;

[0049] Figure 6 This is a flowchart of a user topic tag generation method based on a preferred embodiment of the present application for enterprise WeChat discussion groups;

[0050] Figure 7 This is a structural block diagram of a user topic tag generation system based on enterprise WeChat discussion groups according to an embodiment of this application;

[0051] Figure 8 This is a preferred structural block diagram of a user topic tag generation system based on enterprise WeChat discussion groups according to an embodiment of this application;

[0052] Figure 9 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present application. Attached image description:

[0054] Word vector training module 1; Priority word acquisition module 2; Session tagging module 3;

[0055] Topic tag generation module 4; Information deduplication unit 41; Low-frequency data cleaning unit 42;

[0056] Topic tag generation unit 43; External word vector filtering unit 11; Word vector output unit 12;

[0057] Session archive scanning unit 31; information output unit 32; processor 81;

[0058] Memory 82; Communication interface 83; Bus 80. Detailed Implementation

[0059] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely used to explain this application and are not intended to limit this application. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts are within the scope of protection of this application.

[0060] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0061] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0062] Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by a person of ordinary skill in the technical field to which this application belongs. The words "one", "a", "the" and the like used in this application do not indicate a limit on quantity and may indicate the singular or plural. The terms "include", "comprise", "have" and any variations thereof used in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units that are inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The word "multiple" used in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0063] This embodiment provides a method for generating user topic tags based on enterprise WeChat discussion groups. Figure 1 This is a flowchart of a user topic tag generation method based on WeChat Work discussion groups according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps:

[0064] In the word vector training step S1, different discussion tags are preset according to the industry type, and corresponding external word vectors are obtained by filtering them. The external word vectors are combined with the conversation archive to train word vectors.

[0065] In the priority word acquisition step S2, the similarity between the word vector and the discussion tag is calculated, and the word vectors that meet the preset conditions are added to the priority word list;

[0066] In the conversation tagging step S3, based on priority words and preset stop words, the word segmentation system is used to scan the conversation archive and tag the conversation archive.

[0067] Step S4, which generates topic tags, involves processing the tagging results to obtain user topic tags.

[0068] By following the steps above, and utilizing WeChat Work's conversation archives, user tags can be constructed using preset discussion tags and word vectors, which can accelerate the construction of discussion tags in user conversations and save the workload of operations staff in tagging.

[0069] It should be noted that word vector technology can effectively describe the similarity between words, mapping words or phrases in a vocabulary to vectors of real numbers.

[0070] In some of these embodiments, Figure 2 This is a flowchart of the word vector training steps according to an embodiment of this application, such as... Figure 2 As shown, word vector training step S1 also includes:

[0071] The external word vector selection step S11 involves obtaining the Tencent AI Lab Ab word vectors from Tencent AI Lab, calculating the Euclidean distance between the Tencent AI Lab Ab word vectors and the corresponding vectors of the discussion tags, and selecting external word vectors accordingly.

[0072] The Euclidean distance mentioned above is the actual distance between two points in n-dimensional space. Euclidean distance is used to calculate the similarity of images; the smaller the Euclidean distance, the greater the similarity. The specific calculation method is as follows:

[0073] Given two points A = (a1, a2, ..., a... n B = (b1, b2, ..., b) n If ), then the Euclidean distance between A and B is .

[0074]

[0075] In step S12, the conversation archive is preprocessed to obtain the corresponding one-hot vector. The one-hot vector and the external word vector are then input into the Word2vec model and the word vector is output.

[0076] In the above steps, conversation archives are used and external word vectors are used to obtain word vectors. Specifically, Tencent AILAb word vectors are first compared with preset discussion tags to obtain external word vectors that are similar to the discussion tags, which are then used to supplement the word vector training process based on the external word vectors.

[0077] This solves the problem that it is difficult to represent word information in human language by directly using discussion sessions to train word vectors, since individual discussion messages in a discussion session are usually short.

[0078] The external word vector method can also be replaced by other techniques that search for Chinese synonyms or by manual judgment.

[0079] It should be noted that, under normal circumstances, the external word vectors are dense in dimension and can be set to 200 dimensions, but this invention is not limited to this.

[0080] In some embodiments, the priority word acquisition step S2 specifically includes:

[0081] The cosine similarity between the word vector and the word vector corresponding to the discussion tag is calculated using the following formula.

[0082]

[0083] Where A and B represent word vectors and word vectors corresponding to discussion tags, respectively, n represents the total dimension, and i represents the i-th dimension.

[0084] When the cosine similarity is greater than a preset threshold, the word segment corresponding to the word vector is added to the priority word list.

[0085] The similarity between the words corresponding to the word vectors and the discussion tags can also be determined by the Euclidean distance and Hamming distance between the vectors.

[0086] In some of these embodiments, Figure 3 This is a flowchart of the session tagging steps according to an embodiment of this application, such as... Figure 3 As shown, the session tagging step S3 specifically includes:

[0087] In the conversation archive scanning step S31, priority words and preset stop words are added to the jieba word segmentation system, and the jieba word segmentation system is used to scan the conversation archive.

[0088] In the information output step S32, the corresponding word segment in the session archive is deleted according to the preset stop words, and the corresponding priority words, discussion tags and speaking users that are hit in the remaining part of the session archive are output.

[0089] In some of these embodiments, Figure 4 This is a flowchart of the session tagging steps according to an embodiment of this application, such as... Figure 4 As shown, step S4 of generating topic tags specifically includes:

[0090] Information deduplication step S41: After deduplicating the corresponding priority words and speaking users output in the information output step, we obtain the hit priority words and the number of corresponding speaking users.

[0091] In the low-frequency data cleaning step S42, the variance and mean of the number of hit priority words and the number of speaking users are calculated, and the 3*sigma principle is used for filtering.

[0092] In step S43, the topic tag generation process outputs the information of the speaking user-discussion tag based on the filtering results, and obtains the user's topic tag accordingly.

[0093] The above 3sigma principle states that the probability of a value distribution in (μ-3σ, μ+3σ) is 0.9974. In the normal distribution, σ represents the standard deviation and μ represents the mean. x = μ is the axis of symmetry of the image. By using the 3sigma principle to filter out the priority words that are outside (μ-3σ, μ+3σ), information redundancy and interference from low-frequency conversation information can be effectively reduced.

[0094] The aforementioned low-frequency data cleaning steps can also directly use tags as cleaning targets.

[0095] By performing the above steps to clean the obtained keywords, we can filter out some tags with fewer discussions and retain the more popular discussion topic tags.

[0096] In some embodiments, the filtering conditions in the low-frequency data cleaning step S42 further include:

[0097] If the number of users who speak when a priority keyword is hit is less than a set value, the priority keyword will be filtered out. The set value is the mean minus 3 * standard deviation.

[0098] This embodiment also provides a method for generating user topic tags based on enterprise WeChat discussion groups. Figure 5 This is a flowchart of another method for generating user topic tags based on WeChat Work discussion groups according to an embodiment of this application, such as... Figure 5 As shown, the process includes the following steps:

[0099] S501, Stop Word Data Preparation

[0100] In NLP tasks, there are often a large number of useless words, such as "of" and "we". These words are defined as stop words.

[0101] S502, Preset discussion tags

[0102] For user discussion session archives, operators need to preset different discussion tags according to different industries.

[0103] S503, Obtain word vectors

[0104] 1. Using Tencent AI Lab's open-source large-scale Tencent AI Lab word vectors, we searched for possible matching words similar to the tags discussed in S502, obtained the corresponding external word vectors, and supplemented the word vector training process described below accordingly.

[0105] In this process, the similarity between words can be described by the Euclidean distance between vectors.

[0106] 2. Preprocess the conversation archive. Generally, the preprocessing steps are related to the type of conversation archive and personal purpose. For example, if it is an English conversation, it may be necessary to perform operations such as capitalization and spelling error checking. If it is a Chinese or Japanese conversation, word segmentation processing is required.

[0107] 3. After processing, the processed session archive is obtained. The one-hot vector of the processed session archive and the external word vectors are used as input to the word2vec model. The word2vec model is then used to train low-dimensional word embeddings.

[0108] In practical applications, the above steps can be performed using two training models (CBOW and Skip-gram) and two acceleration algorithms (Negative Sample and Hierarchical Softmax).

[0109] S504, Add priority words

[0110] Through S502 and S503, word vectors are used to calculate words whose cosine similarity to the word vectors of preset discussion tags is greater than a threshold α. These words are then added to the priority word list for word segmentation. Words with similar similarity are assigned to represent the preset topic tags. That is, a topic tag can correspond to multiple sets of words. The cosine similarity is shown below:

[0111]

[0112] Where A and B represent word vectors corresponding to different words. n represents the total dimensions, which is 200 dimensions in step three, and i represents the i-th dimension.

[0113] S505, Session Tagging

[0114] In this step, the stop words from S501 and the priority words from S504 are added to the jieba word segmentation system, and the jieba word segmentation system is used to scan the session archive. If a corresponding priority word is hit, relevant information is output and a corresponding table is generated. The relevant information includes the original session, the hit priority word, the hit tag, and the speaker. S506, Data Cleaning

[0115] 1. After obtaining the table of S505, since the topics discussed are usually scattered. For the hit words, after de-duplicating using the speaking users, the information of the hit words + the number of speaking users is obtained.

[0116] 2. For all the hit words and the number of users, the variance u and the mean sigma are statistically calculated. After obtaining the variance and the mean, the 3*sigma principle is used for information filtering, and the hit words outside (μ - 3σ, μ + 3σ) are filtered.

[0117] 3. If the number of users of a hit word is less than the mean - 3*standard deviation, then the discussion of this hit word is less. And for the table of S505, if its hit words are not in the hit words of S506, then they are filtered.

[0118] S507, User Session Label Construction

[0119] After obtaining the table of S506, the information of speaking users - hit labels is output, and based on this, user session labels are constructed to assist in the construction of downstream tasks such as subsequent user marketing and event planning.

[0120] The following describes and illustrates the embodiments of the present application through preferred embodiments.

[0121] Figure 6 It is a flowchart of a method for generating user topic labels based on enterprise WeChat discussion groups according to the preferred embodiments of the present application.

[0122] S601, Stopword Data Preparation and Adding to the Word Segmentation System

[0123] In NLP tasks, there are usually a large number of useless words, such as words like "of, we", etc. Such words are defined as stopwords and added to the word segmentation system to delete the relevant words in the conversation archive.

[0124] S602, Presetting Discussion Labels in the Beauty Field

[0125] In some beauty fields, the daily activity information and beauty information of users may be concerned. Based on this, the preset discussion labels can be as follows:

[0126] '上班','加班','恋爱','上学','逛街','旅行','在家','美白','口红','游戏','彩妆','护肤'

[0127] S603, Training Word Vectors

[0128] 1. Use Tencent AI Lab's open-source large-scale Tencent AI Lab word vectors to find possible matching words similar to the tags discussed in S502, obtain the corresponding external word vectors, and supplement the following word vector training process accordingly. Typically, the external word vectors are dense in dimension and can be set to 200 dimensions.

[0129] In this process, the similarity between words can be described by the Euclidean distance between vectors.

[0130] 2. Preprocess the session archive.

[0131] 3. After processing, the processed session archive is obtained. The one-hot vector of the processed session archive and the external word vectors are used as input to the word2vec model. The word2vec model is then used to train low-dimensional word embeddings.

[0132] S604, Add priority keywords for the beauty category.

[0133] Using S602 and S603, words with a cosine similarity greater than a threshold 'a' to the word vectors of preset discussion tags are added to the priority word list for word segmentation. Words with similar similarity are then assigned to represent the preset topic. Typically, 'a' is set to 0.6, and the cosine similarity is shown below:

[0134]

[0135] Where A and B represent word vectors corresponding to different words. n represents the total dimensions, which is 200 dimensions in S603, and i represents the i-th dimension.

[0136] S605, Tagging Session Archives

[0137] In this step, stop words from S601 and priority words from S604 are added to the jieba word segmentation system, and the jieba word segmentation system is used to scan the session archive. If a corresponding priority word is matched, the relevant information is output and a table is generated. An example table is shown below:

[0138]

[0139] S606 cleans the output data of S605.

[0140] After obtaining the S605 table, since the discussion topics are usually scattered, for each matched word, deduplication is performed using the speaking users to obtain information on the matched word plus the number of speaking users. For example, lipstick + 2 matched users.

[0141] For all matched words and the number of users, the variance u and mean sigma are calculated. After obtaining the variance and mean, the 3*sigma principle is used to filter the information, obtaining relevant information within (μ-3σ, μ+3σ) and generating a table.

[0142] If the number of users who matched a word is less than the mean - 3 * standard deviation, then the matched word has been discussed less and should be filtered out.

[0143] If the matched words in the table generated by S605 are not among the matched words in the table generated by S606, then filtering is performed.

[0144] S607, Building User Session Tags

[0145] After obtaining the S606 table, output the information of the speaking users and the hit tags to assist in the construction of subsequent downstream tasks such as user marketing and event planning.

[0146] It should be noted that the steps shown in the above process or in the flowchart of the accompanying figures can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0147] This embodiment also provides a user topic tag generation system based on WeChat Work discussion groups. This device is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the terms "module," "unit," "subunit," etc., can refer to a combination of software and / or hardware that performs a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0148] Figure 7 This is a structural block diagram of a user topic tag generation system based on enterprise WeChat discussion groups according to an embodiment of this application, such as... Figure 7 As shown, the device includes:

[0149] The word vector training module 1 presets different discussion tags according to industry type and selects corresponding external word vectors accordingly. The external word vectors are combined with the conversation archive to train word vectors.

[0150] Priority word acquisition module 2 calculates the similarity between word vectors and discussion tags, and adds word vectors that meet preset conditions to the priority word list;

[0151] Module 3 for conversation tagging: The conversation tagging steps involve scanning the conversation archive using a word segmentation system and tagging the conversation archive based on priority words and preset stop words.

[0152] Topic tag generation module 4 processes the tagging results to obtain user topic tags.

[0153] This application uses NLP and other methods to extract user chat information, which can quickly mine user tags in the session archive, help operators automatically tag users, and filter session information with less discussion.

[0154] In some embodiments, the topic tag generation module 4 includes:

[0155] Information deduplication unit 41 receives the corresponding priority words and speaking users output by the session tagging module and deduplicates them to obtain the hit priority words and the number of corresponding speaking users;

[0156] The low-frequency data cleaning unit 42 calculates the variance and mean of the number of hit priority words and the number of speaking users, and filters them using the 3*sigma principle;

[0157] The topic tag generation unit 43 outputs the information of the speaking user-discussion tag based on the filtering results, and obtains the user's topic tag accordingly.

[0158] Figure 8 This is a preferred structural block diagram of a user topic tag generation system based on enterprise WeChat discussion groups according to an embodiment of this application, such as... Figure 8 As shown, the device includes Figure 7 All modules shown, in addition to:

[0159] Word vector training module 1 also includes:

[0160] The external word vector filtering unit 11 calculates the Euclidean distance between the Tencent AI Lab Ab word vectors obtained from Tencent AI Lab and the corresponding vectors of the discussion tags, and filters out external word vectors accordingly.

[0161] The word vector output unit 12 preprocesses the conversation archive to obtain the corresponding one-hot vector, inputs the one-hot vector and external word vectors into the Word2vec model and outputs word vectors.

[0162] The priority word acquisition module 2 calculates the cosine similarity between word vectors and discussion tags using the following formula:

[0163]

[0164] Where A and B represent word vectors and word vectors corresponding to discussion tags, respectively, and n represents the total dimension.

[0165] When the cosine similarity is greater than a preset threshold, the word segment corresponding to the word vector is added to the priority word list.

[0166] Session tagging module 3 specifically includes:

[0167] The conversation archive scanning unit 31 adds priority words and preset stop words to the jieba word segmentation system and uses the jieba word segmentation system to scan the conversation archive.

[0168] Information output unit 32 deletes the corresponding word segment from the session archive according to the preset stop words, and outputs the corresponding priority words, discussion tags and speaking users in the remaining part of the session archive to the information deduplication unit 41.

[0169] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0170] In addition, combined Figure 1 The user topic tag generation method based on Enterprise WeChat discussion groups described in this application can be implemented by a computer device. Figure 9 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present application.

[0171] The computer device may include a processor 81 and a memory 82 storing computer program instructions.

[0172] Specifically, the processor 81 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0173] The memory 82 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 82 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 82 may include removable or non-removable (or fixed) media. Where appropriate, the memory 82 may be internal or external to a data processing device. In a particular embodiment, the memory 82 is non-volatile memory. In a particular embodiment, the memory 82 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0174] The memory 82 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 81.

[0175] The processor 81 reads and executes computer program instructions stored in the memory 82 to implement any of the user topic tag generation methods based on enterprise WeChat discussion groups in the above embodiments.

[0176] In some embodiments, the computer device may further include a communication interface 83 and a bus 80. For example, Figure 8 As shown, the processor 81, memory 82, and communication interface 83 are connected through bus 80 and complete communication with each other.

[0177] The communication interface 83 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of this application. The communication port 83 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0178] Bus 80 includes hardware, software, or both, that couples components of a computer device together. Bus 80 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 80 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 80 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0179] The computer device can perform the session tagging steps in the embodiments of this application based on the acquired session archive, thereby achieving a combination of... Figure 1 This describes a method for generating user topic tags based on WeChat Work discussion groups.

[0180] Furthermore, in conjunction with the user topic tag generation method based on enterprise WeChat discussion groups in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the user topic tag generation methods based on enterprise WeChat discussion groups in the above embodiments.

[0181] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0182] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for generating user topic tags based on WeChat Work discussion groups, characterized in that, Includes the following steps: The word vector training steps involve pre-setting different discussion tags based on industry type, and then filtering them to obtain corresponding external word vectors. These external word vectors are then combined with conversation archives to train the word vectors. The priority word acquisition step involves calculating the similarity between the word vector and the discussion tag, and adding the word vectors that meet the preset conditions to the priority word list; The conversation tagging step involves scanning the conversation archive using a word segmentation system and tagging the conversation archive based on priority words and preset stop words. The topic tag generation process involves processing the tagging results to obtain user topic tags; The word vector training step further includes: The external word vector filtering step involves obtaining Tencent AI Lab Ab word vectors from Tencent AI Lab, calculating the Euclidean distance between the Tencent AI Lab Ab word vectors and the corresponding vectors of the discussion tags, and filtering the external word vectors accordingly. The word vector output step involves preprocessing the session archive to obtain the corresponding one-hot vector, inputting the one-hot vector and the external word vector into the Word2vec model, and outputting the word vector. The topic tag generation step specifically includes: The information deduplication step involves deduplicating the corresponding priority words and speaking users output in the information output step to obtain the number of hit priority words and corresponding speaking users. The low-frequency data cleaning step involves calculating the variance and mean of the hit priority words and the number of speaking users, and then filtering them using the 3*sigma principle. When the number of speaking users corresponding to the hit priority word is less than a set value, the hit priority word is filtered out, where the set value = mean - 3*standard deviation. The topic tag generation process involves outputting the discussion tag information of the speaking user based on the filtering results, and obtaining the user's topic tag accordingly.

2. The user topic tag generation method according to claim 1, characterized in that, The steps for obtaining priority words specifically include: The cosine similarity between the word vector and the tag word vector corresponding to the discussion tag is calculated using the following formula. Where A and B represent the word vector and the word vector corresponding to the discussion tag, respectively, n represents the total dimension, and i represents the i-th dimension. When the cosine similarity is greater than a preset threshold, the word segment corresponding to the word vector is added to the priority word list.

3. The user topic tag generation method according to claim 1, characterized in that, The session tagging steps specifically include: The session archive scanning step involves adding the priority words and the preset stop words to the jieba word segmentation system and using the jieba word segmentation system to scan the session archive. The information output step involves deleting the corresponding word segment from the session archive based on the preset stop words, and outputting the corresponding priority words, discussion tags, and speaking users that are matched in the remaining part of the session archive.

4. A user topic tag generation system based on enterprise WeChat discussion groups, using the user topic tag generation method according to any one of claims 1-3, characterized in that, include: The word vector training module presets different discussion tags according to industry type and selects corresponding external word vectors accordingly. The external word vectors are combined with the conversation archive to train word vectors. The priority word acquisition module calculates the similarity between the word vector and the discussion tag, and adds the word vectors that meet the preset conditions to the priority word list; The conversation tagging module includes the following steps: based on priority words and preset stop words, the conversation archive is scanned using a word segmentation system and the conversation archive is tagged. The topic tag generation module processes the tagging results to obtain user topic tags; The word vector training module includes: The external word vector filtering unit calculates the Euclidean distance between the Tencent AI LAb word vectors obtained from Tencent AI Lab and the corresponding vectors of the discussion tags, and filters out the external word vectors accordingly. The word vector output unit preprocesses the conversation archive to obtain the corresponding one-hot vector, inputs the one-hot vector and the external word vector into the Word2vec model and outputs the word vector. Specifically, the topic tag generation unit includes: The information deduplication unit deduplicates the corresponding priority words and speaking users output in the information output step to obtain the number of hit priority words and corresponding speaking users. The low-frequency data cleaning unit calculates the variance and mean of the hit priority words and the number of speaking users, and filters them using the 3*sigma principle; when the number of speaking users corresponding to the hit priority words is less than a set value, the hit priority words are filtered out, where the set value = mean - 3*standard deviation. The topic tag generation process involves outputting the discussion tag information of the speaking user based on the filtering results, and obtaining the user's topic tag accordingly.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the user topic tag generation method as described in any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the user topic tag generation method as described in any one of claims 1 to 3.

Images