Long text data recognition method, device, electronic device and storage medium
By using the theme aggregation model in resume analysis to segment and annotate long text data, combined with LDA and BERT models, the problem of low recognition accuracy of long text data in the existing technology is solved, and the accuracy of resume analysis and job search is improved. Experience.
Patent Information
- Application Number
- CN202110217134.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-02-26
- Publication Date
- 2024-12-17
- Estimated Expiration
- 2041-02-26
AI Technical Summary
In the analysis of resumes, the recognition effect of long text data such as work experience and project experience is poor, and the named entity recognition model based on neural network has low accuracy in segmentation and labeling of long text, which affects resume analysis. Accuracy.
The long text data is clustered based on a predetermined theme aggregation model, multiple theme module data are generated, and input it to the pre-trained text recognition model for annotation, identify and input it to the corresponding text area, and use LDA theme aggregation The combination of the model and the BERT model is used to segment and annotate content.
It improves the labeling accuracy of the text recognition model, enhances the content segmentation ability of long text data, and improves the accuracy of resume analysis and job seeker delivery experience.
Smart Images
Figure CN113010672B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of big data technology, and in particular to a long text data recognition method, device, electronic device and storage medium. Background Art
[0002] Resume parsing technology has gradually become a standard feature of various recruitment systems. This technology can quickly and accurately parse resume text and import it into the recruitment system, which can improve the job seekers' application experience.
[0003] The resume recognition method based on rules and dictionaries can achieve relatively ideal results for information with obvious characteristics such as graduation schools, email addresses, and mobile phone numbers. However, it is not effective in analyzing content such as work experience, project experience, and personal interests and hobbies in resumes.
[0004] The neural network-based named entity recognition model has some restrictions on the length of the input text sequence. For texts with more words, such as resumes, it is usually necessary to segment the text into modules first and then input them into the model for annotation. However, for texts without obvious delimiters, such as resumes, the commonly used segmentation method is to segment the text according to known feature keywords. However, this method has poor generalization ability and poor segmentation effect. The module segmentation results with large errors will affect the subsequent model's understanding of the overall semantics and affect the annotation results, thereby reducing the accuracy of resume parsing. Summary of the invention
[0005] In view of this, the present invention provides a long text data recognition method, device, electronic device and storage medium to solve at least one of the above-mentioned problems.
[0006] According to a first aspect of the present invention, a long text data recognition method is provided, the method comprising:
[0007] Acquire long text data, wherein the long text data includes: multiple topics;
[0008] Performing clustering processing on the long text data based on a predetermined topic aggregation model to generate a plurality of topic module data;
[0009] Inputting the plurality of subject module data into a pre-trained text recognition model respectively, generating a labeling result for each subject module data, wherein the labeling result is used to identify the subject of the data;
[0010] The topic data of the long text data are identified according to the labeling results of the topic model data, and the identified topic data are input into the corresponding text area part.
[0011] According to a second aspect of the present invention, there is provided a long text data recognition device, the device comprising:
[0012] A data acquisition unit, used to acquire long text data, wherein the long text data includes: a plurality of topics;
[0013] A topic module data generating unit, used for performing clustering processing on the long text data based on a predetermined topic aggregation model to generate a plurality of topic module data;
[0014] A labeling unit, used to input the plurality of subject module data into a pre-trained text recognition model respectively, and generate a labeling result for each subject module data, wherein the labeling result is used to identify the subject of the data;
[0015] An identification unit, used for identifying each topic data of the long text data according to the labeling results of each topic model data;
[0016] The input unit is used to input the recognized subject data into the corresponding text area.
[0017] According to a third aspect of the present invention, there is provided an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.
[0018] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.
[0019] It can be seen from the above technical solution that the acquired long text data is clustered based on a predetermined topic aggregation model to generate multiple topic module data, and then each topic module data is input into a text recognition model to generate a labeling result for each topic module data, and then each topic data of the long text data is identified according to the labeling result of each topic model data, and the identified topic data is input into the corresponding text area part. Compared with the prior art, this technical solution is based on the advantages of the topic aggregation model in text clustering, and can accurately segment the content of long text data, thereby improving the labeling accuracy of the text recognition model, thereby improving the accuracy of recognizing long text data. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0021] Figure 1 is a flowchart of a long text data recognition method according to an embodiment of the present invention;
[0022] Figure 2 is a detailed flow chart of a long text data recognition method according to an embodiment of the present invention;
[0023] Figure 3 is a schematic diagram of the structure of training data according to an embodiment of the present invention;
[0024] Figure 4 is a schematic diagram of the input representation of the BERT text recognition model according to an embodiment of the present invention;
[0025] Figure 5 Schematic diagram of the overall structure of the BERT text recognition model according to an embodiment of the present invention;
[0026] Figure 6 is a structural block diagram of a long text data recognition device according to an embodiment of the present invention;
[0027] Figure 7 It is a schematic block diagram of the system structure of the electronic device 600 according to an embodiment of the present invention. DETAILED DESCRIPTION
[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0029] At present, resumes are recognized through named entity recognition models such as BERT (Bidirectional Encoder Representation from Transformers) and BiLSTM-CRF (a sequence labeling algorithm). Since these models have restrictions on the length of the input sequence, it is necessary to first segment the long text sequence such as the resume and then input it into the model for labeling. The commonly used content segmentation method is to segment the content based on the title keywords of the resume (such as educational experience, work experience), but this method has poor generalization ability. When encountering resumes with uncommon titles or no titles, effective content segmentation will not be performed. The accuracy of content segmentation will affect the model's understanding of the overall context. If the content segmentation accuracy is low, it will lead to a decrease in the model's labeling accuracy, resulting in a low accuracy rate in the model's resume recognition.
[0030] Based on this, an embodiment of the present invention provides a long text data recognition solution, which accurately segments the content of long text data based on the advantages of a predetermined topic aggregation model in text clustering, improves the annotation accuracy of the text recognition model, and thus can improve the accuracy of long text data recognition. The following is a detailed description of the embodiment of the present invention in conjunction with the accompanying drawings.
[0031] Figure 1 is a flow chart of a long text data recognition method according to an embodiment of the present invention. Figure 1 As shown, the method includes:
[0032] Step 101, obtaining long text data, wherein the long text data includes: a plurality of topics.
[0033] The long text data here can be a text with multiple topics and a large number of characters, such as an electronic resume. The length of the text character sequence of the long text data exceeds the character threshold length of the text recognition model input sequence. Therefore, before executing the text recognition model, the long text data needs to be segmented, that is, the following step 102 is executed.
[0034] Step 102: clustering the long text data based on a predetermined topic aggregation model to generate a plurality of topic module data.
[0035] The topic aggregation model here can be a topic aggregation model based on the LDA (Latent Dirichlet Allocation) algorithm. Taking advantage of the LDA topic aggregation model in text clustering, clustering processing (or content segmentation processing) is performed on the long text data, and then the content obtained after the clustering processing is imported into the text recognition model in step 103 for annotation operation.
[0036] Specifically, the long text data may be segmented into lines first; and then each line of segmented data may be segmented according to a predetermined word segmenter.
[0037] Subsequently, a clustering operation is performed on each line of segmented data in the long text data based on the LDA topic aggregation model to determine the topic to which each line of data belongs; and module data of the topic to which it belongs is generated based on each line of data and its topic, thereby generating the multiple topic module data.
[0038] Step 103, respectively input the plurality of subject module data into a pre-trained text recognition model to generate a labeling result for each subject module data, wherein the labeling result is used to identify the subject of the data.
[0039] In one embodiment, when the character length of the subject module data still exceeds the character threshold length of the text recognition model input sequence, the subject module data can be split according to the character threshold length to generate multiple subject sub-module data. Afterwards, each subject sub-module data is input into the text recognition model to generate a labeling result for each subject sub-module data.
[0040] The text recognition model here can be a BERT text recognition model (ie, the BERT named entity recognition model mentioned above).
[0041] In actual operation, the text recognition model can be trained in the following manner: first, historical long text data is obtained, and the historical long text data includes: various topic data with annotation results; then, training data is generated according to the historical long text data, and the text recognition model is trained based on a predetermined loss function (for example, a cross entropy function), a predetermined activation function (for example, a softmax function) and the training data.
[0042] Step 104 , identifying each topic data of the long text data according to the annotation results of each topic model data, and inputting the identified each topic data into the corresponding text area part.
[0043] The acquired long text data is clustered based on a predetermined topic aggregation model to generate multiple topic module data, and then each topic module data is input into a text recognition model to generate a labeling result for each topic module data, and then each topic data of the long text data is identified according to the labeling result of each topic model data, and each identified topic data is input into a corresponding text area part. Compared with the prior art, the embodiment of the present invention can accurately segment the content of the long text data based on the advantage of the topic aggregation model in text clustering, thereby improving the labeling accuracy of the text recognition model, thereby improving the accuracy of recognizing the long text data.
[0044] In order to more clearly describe the present invention, the following takes the long text data as an electronic resume (referred to as resume) as an example and combines Figure 2 The embodiments of the present invention are described in detail.
[0045] like Figure 2 As shown, in this example, the long text data recognition process mainly includes three parts: BERT model construction, resume parsing, and model evolution. The following describes these three parts respectively.
[0046] 1. BERT model construction
[0047] Step 1: Determine the information type and label.
[0048] Generally speaking, the information in a resume can be roughly divided into six categories: basic personal information, education experience, work (internship) experience, language ability, honors and awards, and other information. For example, name, gender, age, contact information, etc. belong to basic personal information. Education experience, major, graduation school, etc. belong to education experience. Work unit, job position, work experience content, etc. belong to work (internship) experience. CET-4 and CET-6 or TOEFL and IELTS scores belong to language ability. Information on various awards received belongs to honors and awards.
[0049] In the embodiment of the present invention, each information is divided into three parts: the beginning, the middle, and the end. For example, each character in the name of the work unit is divided into three parts: the beginning of the work unit, the middle of the work unit, and the end of the work unit. The final label type is shown in Table 1 below:
[0050]
[0051]
[0052] Table 1
[0053] In actual use, each corporate recruitment website can modify the labels according to the type of information required by its own recruitment system.
[0054] Step 2: Generate training data.
[0055] Use the stored historical resume data, that is, the structured historical resume data, and randomly add some meaningless characters (such as spaces, line breaks, etc.) that are commonly seen in actual resume parsing to generate training data for the BERT text recognition model. The structure of the training data is as follows: Figure 3 As shown. Finally, it needs to be converted into a one-hot encoding representation based on all categories to facilitate the calculation of the subsequent cross entropy loss function.
[0056] by Figure 3 For example, the first character "2" input in the corresponding encoding vector has only the position corresponding to category TS as 1, and the values of other positions are all 0. Here, the i-th character in the input sequence X is denoted as x i ,Right now Where M is the maximum length of the input sequence set by the model. When the input sequence length is less than M, placeholders are used to fill the position. i The corresponding label is t i , that is, the output tag sequence is where t i is an n-dimensional vector, where n is the number of label types.
[0057] Step 3: Model construction.
[0058] Figure 4 Schematic diagram of the input representation for the BERT text recognition model, as Figure 4 shown. The original input of the model is a text sequence X, which needs to be converted into word vectors, text vectors, and position vectors. The sum of these three vectors is the input representation B of the BERT network. Word vectors are a combination of the encodings of each word in the text, text vectors reflect the sentences to which each word in the text belongs, and position vectors reflect the positions where each word in the text appears.
[0059] Figure 5 Schematic diagram of the overall structure of the BERT text recognition model, as Figure 5 shown. At the last layer of the BERT network, a feed-forward neural network is connected as a classifier and activated using the softmax function. The final output sequence is where y i is the probability vector for the input character x i belonging to each category (the category with the highest probability is the category to which the character belongs). Taking Figure 5 the first character "中" in the input as an example, the corresponding output sequence y 1 has the label with the highest probability as C-S. So, this character is classified as the C-S category, which is the start of the work unit name.
[0060] In this example, the cross-entropy function can be used as the loss function of the model.
[0061] The loss function is defined as follows:
[0062] where,
[0063] where θ are the network parameters.
[0064] Step 4, according to the model constructed in Steps 1-3, start model training.
[0065] II. Resume Parsing
[0066] Step 1, import the resume file.
[0067] Step 2, content segmentation.
[0068] (1) First, segment the content obtained from the resume by line, with each line as an independent text. For all the texts, perform word segmentation using a word segmenter, and the resulting set of words is the dictionary for the clustering operation. This word segmenter can use the word segmenters in the prior art, and the present invention places no restrictions thereon.
[0069] (2) Determine the number k of topic categories, which can be determined according to the number of major categories in Table 1.
[0070] (3) Use the LDA model to cluster the text and obtain the topic category to which each line of content belongs.
[0071] (4) Based on the clustering results and the order in which the text appears, consecutive lines with the same topic are grouped into the same module.
[0072] Step 3: Input the text into the BERT model for sequence annotation in modules to obtain the classification corresponding to each character. If the character length of the module still exceeds the maximum sequence length accepted by the model, the module is split at the maximum length.
[0073] Step 4: Summarize the resume content according to the classification results output by the BERT model and the order in which the text appears. For example, consecutive sentences marked as work experience are summarized as the same work experience. The content marked as time and work unit that appears in front of the sentence and is closest to it is the time period and work unit corresponding to the work experience. After that, fill the final result into the corresponding input box in the recruitment system to complete the resume parsing.
[0074] 3. Model Evolution
[0075] Step 1: During the resume parsing process, when the user modifies the parsing result, the content in the file stream of the resume is recorded.
[0076] Step 2: Modify the annotation content according to the correct parsing result obtained after the user's correction, and generate a new data set for model evolution training. The specific method is as follows:
[0077] Assume that the original input text sequence is as follows:
[0078] From July 2001 to May 2001, worked in the Software Development Center of Industrial and Commercial Bank of China.
[0079] The label sequence output by the model is as follows:
[0080] TS / TM / TM / TM / TM / TE / TF / TS / TM / TM / TM / TM / TE / N / CS / CM / CM / CM / CM / CE / WE-S / WE-M / WE-M / WE-M / WE-M / WE-M / WE-M / WE-M
[0081] After analyzing and processing the model output results, July 2016-May 2019 will be filled in the time part of the work experience, Industrial and Commercial Bank of China will fill in the work unit module of this work experience, and the work of the Software Development Center will be filled in the description part of the work experience. At this time, the user modifies the work unit to Industrial and Commercial Bank of China Software Development Center. After that, the evolution process of the model is triggered, and the label sequence output by the model will be modified to TS / TM / TM / TM / TM / TE / TF / TS / TM / TM / TM / TM / TE / N / CS / CM / CM / CM / CM / CM / CM / CM / CM / CM / CM / CM / CE / WE-M / WE-M, and then put it into the training data set of model evolution training for evolution training.
[0082] In this way, the annotation accuracy of the text recognition model can be improved, thereby improving the accuracy of recognizing long text data.
[0083] From the above description, it can be seen that the embodiment of the present invention combines the LDA model with the BERT model, utilizes the advantages of the LDA topic aggregation model in text clustering, performs content segmentation on the resume text, and then imports the content obtained after segmentation into the BERT model for sequence labeling operations. In the training stage of the BERT model, a large amount of structured resume data that already exists in the recruitment system can be used to generate training data, and the model can be trained and fine-tuned to finally obtain a labeling model that can be targeted at the resume text. At the same time, in actual use, according to the user's correction of the resume parsing results, the model continues to be trained to further improve the accuracy of the model. Through the embodiment of the present invention, the accuracy of resume parsing in the recruitment system can be improved, and the job seeker's submission experience can be improved.
[0084] Based on similar inventive concepts, an embodiment of the present invention further provides a long text data recognition device. Preferably, the device can be used to implement the process in the above method embodiment.
[0085] Figure 6 is a structural block diagram of the long text data recognition device, such as Figure 6 As shown, the device includes: a data acquisition unit 1, a subject module data generation unit 2, a labeling unit 3, a recognition unit 4 and an input unit 5, wherein:
[0086] The data acquisition unit 1 is used to acquire long text data, wherein the long text data includes: a plurality of topics.
[0087] The topic module data generating unit 2 is used to perform clustering processing on the long text data based on a predetermined topic aggregation model to generate a plurality of topic module data.
[0088] Specifically, the theme module data generation unit includes: a theme determination module and a theme module data generation module, wherein:
[0089] A topic determination module, used to perform a clustering operation on each line of data in the long text data based on the topic aggregation model to determine the topic to which each line of data belongs;
[0090] The subject module data generating module is used to generate the module data of the subject according to each row of data and the subject to which it belongs, so as to generate the plurality of subject module data.
[0091] The labeling unit 3 is used to input the plurality of subject module data into a pre-trained text recognition model respectively, and generate labeling results for each subject module data, wherein the labeling results are used to identify the subject of the data.
[0092] The identification unit 4 is used to identify each topic data of the long text data according to the labeling results of each topic model data.
[0093] The input unit 5 is used to input the recognized subject data into the corresponding text area.
[0094] The long text data acquired by the data acquisition unit is clustered based on a predetermined topic aggregation model by the topic module data generation unit to generate multiple topic module data. Then the annotation unit inputs each topic module data into the text recognition model to generate an annotation result for each topic module data. Subsequently, the recognition unit identifies each topic data of the long text data according to the annotation result of each topic model data. The input unit inputs each identified topic data into the corresponding text area part. Compared with the prior art, the embodiment of the present invention can accurately segment the content of the long text data based on the advantage of the topic aggregation model in text clustering, thereby improving the annotation accuracy of the text recognition model and thus improving the accuracy of the recognition of the long text data.
[0095] In one embodiment, the above device further includes: a segmentation unit and a word segmentation unit, wherein:
[0096] A segmentation unit, used for segmenting the long text data into lines;
[0097] The word segmentation unit is used to perform word segmentation processing on each row of data after segmentation according to a predetermined word segmenter.
[0098] After performing segmentation and word segmentation operations on the long text data, the above-mentioned topic module data generating unit 2 can be executed to perform clustering operations on the long text data.
[0099] In one embodiment, the above-mentioned device also includes: a splitting unit, which is used to split the subject module data according to the character threshold length in response to the character length of the subject module data exceeding the character threshold length of the text recognition model, to generate multiple subject sub-module data.
[0100] In the specific implementation process, the above device also includes: a model training unit for training the text recognition model. Specifically, the model training unit includes: a historical data acquisition module and a model training module, wherein:
[0101] A historical data acquisition module is used to acquire historical long text data, wherein the historical long text data includes: data of each subject with annotated results;
[0102] A model training module is used to generate training data according to the historical long text data, and train the text recognition model based on a predetermined loss function, a predetermined activation function and the training data.
[0103] The specific execution process of the above-mentioned units and modules can be found in the description of the above-mentioned method embodiment, which will not be repeated here.
[0104] In actual operation, the above-mentioned units and modules may be provided in combination or individually, and the present invention is not limited thereto.
[0105] This embodiment also provides an electronic device, which may be a desktop computer, a tablet computer, a mobile terminal, etc., but this embodiment is not limited thereto. In this embodiment, the electronic device may be implemented with reference to the above method embodiment and the embodiment of the long text data recognition device, the contents of which are incorporated herein, and the repeated parts are not repeated.
[0106] Figure 7 FIG. 6 is a schematic block diagram of a system structure of an electronic device 600 according to an embodiment of the present invention. Figure 7 As shown, the electronic device 600 may include a central processor 100 and a memory 140; the memory 140 is coupled to the central processor 100. It should be noted that this figure is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.
[0107] In one embodiment, the long text data recognition function may be integrated into the central processing unit 100. The central processing unit 100 may be configured to perform the following control:
[0108] Acquire long text data, wherein the long text data includes: multiple topics;
[0109] Performing clustering processing on the long text data based on a predetermined topic aggregation model to generate a plurality of topic module data;
[0110] Inputting the plurality of subject module data into a pre-trained text recognition model respectively, generating a labeling result for each subject module data, wherein the labeling result is used to identify the subject of the data;
[0111] The topic data of the long text data are identified according to the labeling results of the topic model data, and the identified topic data are input into the corresponding text area part.
[0112] From the above description, it can be seen that the electronic device provided in the embodiment of the present application performs clustering processing on the acquired long text data based on a predetermined topic aggregation model to generate multiple topic module data, and then inputs each topic module data into a text recognition model to generate a labeling result for each topic module data, and then identifies each topic data of the long text data according to the labeling result of each topic model data, and inputs each identified topic data into the corresponding text area part. Compared with the prior art, the embodiment of the present invention can accurately segment the content of long text data based on the advantages of the topic aggregation model in text clustering, thereby improving the labeling accuracy of the text recognition model, thereby improving the accuracy of recognizing long text data.
[0113] In another embodiment, the long text data recognition device may be configured separately from the central processor 100. For example, the long text data recognition device may be configured as a chip connected to the central processor 100, and the long text data recognition function is realized under the control of the central processor.
[0114] like Figure 7 As shown, the electronic device 600 may further include: a communication module 110, an input unit 120, an audio processor 130, a display 160, and a power supply 170. It is worth noting that the electronic device 600 does not necessarily have to include Figure 7 In addition, the electronic device 600 may also include Figure 7 For components not shown, reference may be made to the prior art.
[0115] like Figure 7 As shown, the central processor 100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor devices and / or logic devices. The central processor 100 receives inputs and controls the operations of various components of the electronic device 600.
[0116] The memory 140 may be, for example, one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory or other suitable devices. The above-mentioned information related to the failure may be stored, and a program for executing the relevant information may also be stored. The CPU 100 may execute the program stored in the memory 140 to implement information storage or processing.
[0117] The input unit 120 provides input to the CPU 100. The input unit 120 is, for example, a key or a touch input device. The power supply 170 is used to provide power to the electronic device 600. The display 160 is used to display display objects such as images and text. The display may be, for example, an LCD display, but is not limited thereto.
[0118] The memory 140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be a memory that saves information even when the power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROMs, etc. The memory 140 may also be some other type of device. The memory 140 includes a buffer memory 141 (sometimes referred to as a buffer). The memory 140 may include an application / function storage unit 142, which is used to store application programs and function programs or processes for executing the operation of the electronic device 600 through the central processor 100.
[0119] The memory 140 may also include a data storage unit 143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 144 of the memory 140 may include various drivers for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).
[0120] The communication module 110 is a transmitter / receiver that transmits and receives signals via the antenna 111. The communication module 110 is coupled to the central processor 100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.
[0121] Based on different communication technologies, multiple communication modules 110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless LAN module. The communication module (transmitter / receiver) 110 is also coupled to a speaker 131 and a microphone 132 via an audio processor 130 to provide an audio output via the speaker 131 and receive an audio input from the microphone 132, thereby realizing a common telecommunication function. The audio processor 130 may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 130 is also coupled to the central processor 100, so that the sound can be recorded on the local machine through the microphone 132, and the sound stored on the local machine can be played through the speaker 131.
[0122] The embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the long text data recognition method are implemented.
[0123] In summary, the embodiment of the present invention combines the LDA model with the BERT model for resume parsing, overcoming the problem that the sequence cannot be input due to excessive length when performing resume parsing in the neural network-based named entity recognition method, and the problem that the model annotation accuracy decreases due to improper content segmentation. Through the embodiment of the present invention, the accuracy of resume parsing in the recruitment system can be improved, and the application experience of job seekers can be improved.
[0124] The preferred embodiments of the present invention are described above with reference to the accompanying drawings. Many features and advantages of these embodiments are clear from this detailed description, and therefore the claims are intended to cover all such features and advantages of these embodiments that fall within their true spirit and scope. In addition, since many modifications and changes are easily conceivable to those skilled in the art, it is not intended to limit the embodiments of the present invention to the precise structure and operation illustrated and described, but all suitable modifications and equivalents falling within its scope may be covered.
[0125] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0126] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0127] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0128] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0129] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.
Claims
1. A long text data recognition method, characterized in that: The method comprises: Acquire long text data, wherein the long text data includes: multiple topics; Performing clustering processing on the long text data based on a predetermined topic aggregation model to generate a plurality of topic module data; Inputting the plurality of subject module data into a pre-trained text recognition model respectively, generating a labeling result for each subject module data, wherein the labeling result is used to identify the subject of the data; The subject data of the long text data are identified according to the marking results of the subject module data, and the identified subject data are input into the corresponding text area part.
2. The method according to claim 1, characterized in that Clustering the long text data based on a predetermined topic aggregation model to generate multiple topic module data includes: Performing a clustering operation on each line of data in the long text data based on the topic aggregation model to determine the topic to which each line of data belongs; The module data of the theme to which each row of data belongs is generated according to the theme to which it belongs, thereby generating the plurality of theme module data.
3. The method according to claim 2, characterized in that Before performing a clustering operation on each line of data in the long text data based on the topic aggregation model, the method further includes: Segmenting the long text data into lines; Each row of data after segmentation is segmented according to the predetermined word segmenter.
4. The method according to claim 1, characterized in that In response to the character length of the subject module data exceeding the character threshold length of the text recognition model, the method further includes: The subject module data is split according to the character threshold length to generate a plurality of subject sub-module data.
5. The method according to claim 1, characterized in that The text recognition model is trained as follows: Acquire historical long text data, wherein the historical long text data includes: data of each subject with annotation results; Training data is generated according to the historical long text data, and the text recognition model is trained based on a predetermined loss function, a predetermined activation function and the training data.
6. A long text data recognition device, characterized in that: The device comprises: A data acquisition unit, used to acquire long text data, wherein the long text data includes: a plurality of topics; A topic module data generating unit, used for performing clustering processing on the long text data based on a predetermined topic aggregation model to generate a plurality of topic module data; A labeling unit, used to input the plurality of subject module data into a pre-trained text recognition model respectively, and generate a labeling result for each subject module data, wherein the labeling result is used to identify the subject of the data; An identification unit, used for identifying each subject data of the long text data according to the marking results of each subject module data; The input unit is used to input the recognized subject data into the corresponding text area.
7. The device according to claim 6, characterized in that The subject module data generating unit comprises: A topic determination module, used to perform a clustering operation on each line of data in the long text data based on the topic aggregation model to determine the topic to which each line of data belongs; The subject module data generating module is used to generate the module data of the subject according to each row of data and the subject to which it belongs, so as to generate the plurality of subject module data.
8. The device according to claim 7, characterized in that The device also includes: A segmentation unit, used for segmenting the long text data into lines; The word segmentation unit is used to perform word segmentation processing on each row of data after segmentation according to a predetermined word segmenter.
9. The device according to claim 6, characterized in that The device also includes: The splitting unit is used to split the subject module data according to the character threshold length to generate multiple subject sub-module data in response to the character length of the subject module data exceeding the character threshold length of the text recognition model.
10. The device according to claim 6, characterized in that The device further comprises: a model training unit, configured to train the text recognition model. The model training unit comprises: A historical data acquisition module is used to acquire historical long text data, wherein the historical long text data includes: data of each subject with annotated results; A model training module is used to generate training data according to the historical long text data, and train the text recognition model based on a predetermined loss function, a predetermined activation function and the training data.
11. An electronic 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 program, the steps of the method according to any one of claims 1 to 5 are implemented.
12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.