System and method for automatically analyzing query request submitted by job seeker or recruiter

Abstract

The invention discloses a system and method for automatically analyzing a query request submitted by a job seeker or a recruiter, and belongs to the technical field of data processing. The method comprises the following steps: receiving a query request submitted by a user, splicing a query statement and user portrait information to form an input sequence, and performing deep semantic analysis on the input sequence by using a deep learning model to generate an embedded vector; the system adopts a prefix array to quickly calculate the sum of word vector intervals, confirms boundaries of words in combination with a word segmentation tool, and generates embedded representation of complete words. And through a multi-intention classification module, performing multi-classification on the embedded vector of the non-MASK position by utilizing a Softmax layer, identifying multiple intentions in the query and outputting corresponding category labels. According to the method, deep semantic understanding, multi-intention separation and personalized recommendation can be realized, and efficient and accurate service support is provided for a recruitment platform and job seekers.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular, relates to a system and method for automatically parsing a query request submitted by a job seeker or a recruiter. Background Art

[0002] In the modern recruitment field, with the rapid growth of job information and the number of job seekers, how to efficiently and accurately parse the query requests submitted by job seekers or recruiters has become the key to improving the efficiency of person-job matching.

[0003] Issues facing the modern recruitment field include:

[0004] - Ambiguous query intent: User search queries often contain abbreviations, industry terms, or non-standard terms, which makes it difficult to understand the intent. For example, a user may search for "front-end development" but does not clearly indicate whether it refers to web front-end or mobile application front-end.

[0005] -Multiple intents: A single query may contain multiple recruitment requirements or job-seeking intentions, which are difficult to distinguish effectively by traditional parsing methods. For example, a user may search for "data analyst" and "data scientist" in the same query, hoping to find both types of positions.

[0006] - Insufficient personalized matching: The matching requirements of job seekers’ preferences, skills and expertise with corporate culture are complex and cannot be fully met by existing technologies. Different job seekers have different expectations for work environment, job requirements or company culture, and traditional methods cannot accurately capture these personalized needs.

[0007] - Information overload processing: How to quickly extract key information and reduce information noise from the massive amount of job information and resumes is another major challenge. User search queries contain a lot of information, and how to find the most relevant jobs or resumes is the core task of recruitment platforms.

[0008] The main existing technologies include:

[0009] Keyword matching: Preliminary screening based on keyword extraction, but it is easy to miss or misjudge due to insufficient semantic understanding. For example, "Java development engineer" is found through keyword matching, but it does not take into account that the user may need a developer with specific industry experience. Rule engine: Filter information according to preset rules, but it is difficult to flexibly respond to changing query needs and personalized preferences. For example, the rule engine may set a rule that "applicants must have 3 years of work experience", but it does not take into account that some positions may require 5 years or more of experience. Initial exploration of NLP: Apply natural language processing technology for semantic understanding, but the depth and breadth still need to be improved. For example, NLP can understand the keywords in user queries, but its in-depth understanding of user intent and contextual analysis capabilities are limited.

[0010] However, the existing technology still has many problems:

[0011] For example: Shallow semantic understanding: Insufficient deep semantic understanding of complex query statements, making it difficult to accurately grasp the user's true intention. For example, if a user queries "needs a copy editor who understands SEO", traditional NLP may only understand "SEO" and "copy editor", but fails to understand that the user needs someone who is good at both SEO skills and content creation.

[0012] Poor dynamic adaptability: The technology is updated slowly, making it difficult to quickly adapt to new demands and new terms in the recruitment market. For example, new technical terms such as "blockchain development" may not be recognized in the old parsing system.

[0013] Lack of personalized analysis: The lack of dynamic analysis strategies for users’ personalized needs leads to low matching accuracy. For example, when matching the personalized needs of different users, traditional algorithms may not be able to distinguish the specific requirements of users for positions.

[0014] Weak multi-intent processing capabilities: When processing queries containing multiple intents, only single or generalized results can be provided. For example, a user may search for "front-end development" and "back-end development" in the same query, and the existing system may only provide job information for one of them. Summary of the invention

[0015] Problem to be solved

[0016] Based on the above reasons, the present invention aims to solve the following technical problems:

[0017] 1. Deep semantic analysis: Through deep learning and advanced NLP technology, we can deeply understand the intent and context behind the query. For example, using the BERT model for deep semantic analysis can better understand the implicit intent in user queries.

[0018] 2. Intelligent separation of multiple intents: Develop algorithm models to effectively identify and separate multiple intents in queries and achieve accurate responses. For example, design a multi-classification model to decompose users' complex queries into multiple single intents.

[0019] 3. Personalized matching optimization: Combine user portraits and behavior analysis to optimize query parsing strategies and improve personalized matching results. For example, dynamically adjust the weight of the matching algorithm based on the user's historical behavior and preferences.

[0020] Deep semantic analysis: Through deep learning and advanced natural language processing (NLP) technology, we can deeply understand the intention and context behind the query requests submitted by users. Traditional methods often only stay at the surface level when understanding the true intention of users and cannot accurately grasp the needs of users. Specifically, the present invention aims to solve: using pre-trained models (such as BERT) to vectorize queries and capture semantic information in queries; calculating the interval sum of tokens through prefix arrays (preSum) to obtain word vector representation, and further improve the accuracy of semantic analysis.

[0021] Intelligent separation of multiple intents: Develop an algorithm model to effectively identify and separate multiple intents in queries to achieve accurate responses. A user's search query may contain multiple recruitment requirements or job-seeking intentions, and traditional parsing methods have difficulty effectively distinguishing these multiple intents. Specifically, the present invention solves this problem in the following ways: Use a word segmentation tool to identify word boundaries, and multi-classify the vector of each non-MASK position through a softmax layer to achieve recognition and separation of multiple intents in the query.

[0022] Personalized matching optimization: Combine user portraits with behavior analysis to optimize query parsing strategies and improve personalized matching effects. The matching requirements of job seekers’ preferences, skills and expertise with corporate culture are complex, and it is difficult for existing technologies to fully meet them. Specifically, the present invention obtains richer user portrait information by: splicing the user’s query statement with the user portrait information (associated with the user’s industry based on the user’s userid) through special characters. During the query parsing process, the weight of the matching algorithm is dynamically adjusted, and personalized matching optimization is performed based on the user’s historical behavior and preferences.

[0023] Technical Solution

[0024] To solve the above problems, the present invention adopts the following technical solutions.

[0025] A system that automatically parses query requests submitted by job seekers or recruiters.

[0026] The data input module is used to receive the query request submitted by the job seeker or the recruiter, and concatenate the query statement and the user profile information through the special character [seq] to form an input sequence;

[0027] The semantic parsing module performs deep semantic parsing on the input sequence based on the Transformer Encoder model, decomposes the query statement into several basic unit tokens, and generates corresponding embedding vectors;

[0028] The prefix array processing module calculates the interval sum of word vectors through the prefix array preSum, including:

[0029] Confirm word boundaries using word segmentation tools;

[0030] The word vector of any interval [i,j] is quickly calculated using the formula preSum[j]-preSum[i-1], and the vectors in the interval are averaged to generate the embedding representation of the complete word;

[0031] The multi-intent classification module inputs the generated embedding vector of each non-MASK position into the Softmax layer for multi-classification to identify multiple intents in the query request and output the corresponding category labels;

[0032] The personalized matching module combines user portrait information and historical behavior data to dynamically adjust the weight of classification results, optimize matching strategies, and generate personalized recommendation results that meet user needs.

[0033] Preferably, the semantic parsing module implements deep semantic parsing by the following steps:

[0034] The user's query statement and user profile information are concatenated using the special character [seq] to form an input sequence;

[0035] Use Transformer Encoder to vectorize the input sequence and generate an embedding vector for each basic unit token;

[0036] The word vector sum of any interval is calculated through the prefix array preSum, including:

[0037] After confirming the word boundary, use the formula preSum[j]-preSum[i-1] to quickly calculate the word vector of the interval [i,j];

[0038] Average the vectors in the interval to generate an embedding representation of the complete word;

[0039] Combined with context information, the self-attention mechanism is used to capture the dependencies between words in the query sentence, thus enhancing the context-awareness of semantic representation.

[0040] Position encoding is introduced to incorporate word order information into word embedding, so that the model can capture the word order structure characteristics.

[0041] Preferably, the multi-intent classification module realizes multi-intent recognition by the following steps:

[0042] The embedding vector of each non-MAS position generated by the semantic parsing module is input into the Softmax layer for multi-classification;

[0043] The Softmax layer outputs the corresponding category labels, which include but are not limited to "position", "industry", "skill", and "language", which are used to identify multiple intents in the query request;

[0044] Separate the multiple identified intents and generate independent parsing results for each intent to achieve hierarchical identification of query intent;

[0045] Combined with user portrait information, the weight of the classification results is dynamically adjusted to ensure that the classification results are highly relevant to user needs.

[0046] Preferably, the personalized matching module implements personalized recommendation through the following steps:

[0047] Obtain user query requests and user profile information, including but not limited to the user's industry, job preferences, and skills and expertise;

[0048] According to the category labels output by the multi-intent classification module, the weights of each category are dynamically adjusted to prioritize matching categories with higher relevance to user profile information;

[0049] Use user historical behavior data, including browsing history and click habits, to build a user interest model and further optimize the recommendation strategy;

[0050] Combine the adjusted category weights and user interest model to generate personalized recommendation results that meet user needs;

[0051] The generated personalized recommendation results are sorted and filtered, and the results with the highest confidence are selected and output to the user end.

[0052] Preferably, the prefix array processing module achieves efficient calculation by the following steps:

[0053] Construct the prefix array preSum: preSum[j] represents the cumulative vector sum from the start of the sequence to position j;

[0054] Quickly calculate the cumulative sum of any interval [i,j]: Use the formula preSum[j]-preSum[i-1] to calculate the cumulative sum of all token embedding vectors in the interval [i,j] in O(1) time;

[0055] Generate word embedding representation: average the cumulative vectors in the interval [i, j] to get the embedding representation of the corresponding complete word;

[0056] Use word segmentation tools to confirm word boundaries: Use word segmentation to accurately locate the start and end positions of each word, ensuring that the interval calculation results accurately correspond to the semantic information of the word.

[0057] Preferably, the system further includes a user portrait enhancement module, which is used to optimize the user portrait information in the following ways:

[0058] Dynamically update user profiles based on historical user behavior data;

[0059] The user profile information and the query request are concatenated using the special character [seq] to form a more comprehensive input sequence;

[0060] In the process of semantic analysis and multi-intent classification, user profile information is used as context reference to improve the ability to understand query intent;

[0061] Adjust the weights of the industry, position, and skill attributes in the user portrait to adapt to personalized needs in different scenarios.

[0062] Preferably, the system further includes a multi-label mapping module for labeling the intent categories in the query request, specifically including:

[0063] Map the intent category in the query request with a predefined tag set, which includes but is not limited to "position", "rank", "industry", "product", "major", "company", "school", "language", and "skill";

[0064] Assign a unique label number to each intent category and associate it with the output of the multi-intent classification module;

[0065] Supports the output of query parsing results with multiple labels to meet the needs of expressing multiple intentions in user queries;

[0066] Through the mapping relationship between label numbers and classification results, the processing efficiency of subsequent personalized recommendation modules is optimized.

[0067] Preferably, the system further includes a label priority sorting module, which is used to prioritize the category labels output by the multi-intent classification module, specifically including:

[0068] Assign initial weights to category labels based on user profile information;

[0069] Dynamically adjust the priority of category labels based on the user's historical behavior data;

[0070] Sort the multiple category labels output by the multi-intent classification module from high to low priority;

[0071] The sorted priority labels are used as input to the personalized recommendation module to optimize the relevance and accuracy of the recommendation results.

[0072] The method for automatically parsing a query request submitted by a job seeker or a recruiter as described above comprises the following steps:

[0073] Receiving query requests: receiving query requests submitted by job seekers or recruiters, and concatenating the query statements and user profile information through special characters to form an input sequence;

[0074] Deep semantic parsing: Use deep learning models to perform deep semantic parsing on the input sequence, decompose the query statement into multiple basic units, and generate corresponding embedding vectors;

[0075] Multi-intent classification: The embedding vector generated by the semantic parsing module is input into the multi-intent classification module to perform multi-classification on the input, identify multiple intents in the query request and output the corresponding category labels;

[0076] Dynamic update of user portraits: Dynamically update user portrait information based on the user's historical behavior data to reflect the user's latest needs and preferences;

[0077] Personalized recommendation generation: Combine the updated user portrait information and multi-intent classification results, adjust the weights of each category, generate personalized recommendation results that meet user needs, sort and filter the results, and output them to the user end.

[0078] Beneficial Effects

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] For the systems involved:

[0081] (1) Significant improvement in deep semantic understanding

[0082] By introducing deep learning models (such as BERT) and advanced natural language processing technology, the present invention can more deeply analyze the semantic information of user query requests. By vectorizing the query statements and combining them with contextual semantic analysis, the real intention of the user can be accurately captured, solving the problem of shallow semantic understanding in traditional methods.

[0083] (2) Enhanced multi-intention processing capabilities

[0084] The present invention can effectively identify and separate multiple intents in a query by designing an intelligent multi-intent separation algorithm. The word segmentation tool and the softmax multi-classification model are used to refine complex queries, achieve accurate responses to multi-intent queries, and overcome the shortcomings of the prior art in multi-intent processing capabilities.

[0085] (3) Optimization of personalized matching effect

[0086] The present invention combines user portraits with behavior analysis, dynamically associates user query requests with their historical behaviors and industry preference information, and optimizes matching strategies through personalized weight adjustment. Compared with traditional rule-based or keyword matching methods, the present invention can provide personalized recommendation results that better meet user needs, greatly improving matching accuracy.

[0087] (4) Strong dynamic adaptability

[0088] The present invention uses a deep learning model to train data, which can quickly adapt to new demands, new terms and industry changes in the recruitment market. Compared with traditional rule engines, the present invention does not require frequent manual rule updates and has higher flexibility and dynamic adaptability.

[0089] (5) Improved information processing efficiency

[0090] The present invention calculates the interval sum of the token embedding vector by introducing a prefix array (preSum) and confirms the word boundary in combination with a word segmentation tool, thereby achieving efficient semantic parsing and information extraction. The method can quickly process massive job information and job resumes, reduce information noise, and improve system response speed.

[0091] (6) Wide range of application scenarios

[0092] Through the above technical solution, the present invention not only improves the automatic parsing capability of query requests, but also greatly enhances the personalized matching effect, providing a more efficient and accurate service experience for recruitment platforms and job seekers.

[0093] At the same time, for the methods involved, improve the matching accuracy: by dynamically updating user portraits and multi-intent classification, it is possible to more accurately match the needs of job seekers and recruiters, thereby improving the relevance and satisfaction of recommendations. Strengthen semantic understanding: the combination of deep semantic parsing and multi-intent recognition enables the system to effectively understand multiple intents when parsing complex query requests, thereby improving the intelligence of the system. Enhance user experience: the generation process of personalized recommendation results takes into account the actual needs of users, and can provide users with services that are more in line with their preferences, thereby improving the overall user experience. Strong adaptability: the dynamic adjustment mechanism in the method enables the system to quickly adapt to changes in the recruitment market, update recommendation strategies in a timely manner, and effectively respond to industry changes.

[0094] The present invention is not only applicable to the matching of job information and job-seeking needs in recruitment platforms, but can also be extended to other fields, such as education and training, e-commerce search, and scenarios that require multi-intent analysis and personalized recommendations, and has a wide range of practical value. In summary, the present invention is significantly superior to the existing technology in terms of semantic understanding, multi-intent processing, personalized optimization, dynamic adaptability, and information processing efficiency, providing a more efficient, accurate, and intelligent service experience for recruitment platforms and job seekers. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 A system structure diagram for automatically parsing query requests submitted by job seekers or recruiters of the present invention;

[0096] Figure 2 It is a system flow architecture diagram of the present invention.

[0097] Figure 3 A technical method diagram of using "prefix sum array (preSum)" to calculate the interval sum of word vectors in a query statement of the present invention;

[0098] Figure 4 This is a flow chart of the method for automatically parsing a query request submitted by a job seeker or a recruiter according to the present invention. DETAILED DESCRIPTION

[0099] The present invention is further described below in conjunction with specific embodiments.

[0100] Example

[0101] The present invention provides a system for automatically parsing a query request submitted by a job seeker or a recruiter.

[0102] like Figure 1 As shown, the present invention provides a system for automatically parsing a query request submitted by a job seeker or a recruiter, comprising the following:

[0103] The data input module is used to receive query requests submitted by job seekers or recruiters, and concatenate the query statements and user profile information through special characters [seq] to form an input sequence; the semantic parsing module performs deep semantic parsing on the input sequence based on the TransformerEncoder model, decomposes the query statements into several basic units token, and generates corresponding embedding vectors; the prefix array processing module calculates the interval sum of word vectors through the prefix array preSum, including: confirming the word boundaries according to the word segmentation tool; quickly calculating the word vector of any interval [i,j] through the formula preSum[j]-preSum[i-1], and averaging the vectors in the interval to generate an embedding representation of the complete word; the multi-intention classification module inputs the generated embedding vector of each non-MASK position into the Softmax layer for multi-classification, which is used to identify multiple intentions in the query request and output the corresponding category label; the personalized matching module dynamically adjusts the weight of the classification results based on user profile information and historical behavior data, optimizes the matching strategy, and generates personalized recommendation results that meet user needs.

[0104] The query parsing system based on Transformer Encoder, combined with prefix array, word segmentation tool and multi-intent classification module, completes the semantic parsing and classification of query requests submitted by job seekers or recruiters.

[0105] Reference Figure 2 , the following is a detailed analysis of each part:

[0106] Input Layer:

[0107] - The user's input query statement (such as "obstetrician-gynecologist") is decomposed into basic units (tokens), such as "woman", "delivery", "department", "doctor", "student".

[0108] - The query statement and the user profile information are concatenated through the special character `[seq]` to form an input sequence, which is used to enrich the context information (such as industry, position preference, etc.).

[0109] Transformer Encoder:

[0110] - The input sequence is fed into the Transformer Encoder model to generate the embedding vectors of each token, such as "woman emb", "delivery emb", etc.

[0111] - These embedding vectors retain the semantic information of each token and provide a basis for subsequent processing.

[0112] Prefix array and word segmentation tool:

[0113] - Prefix array (preSum): Used to quickly calculate the interval sum of word vectors:

[0114] - The interval sum formula is:

[0115] $$\text{Interval sum} = \text{preSum}[j] - \text{preSum}[i - 1]$$

[0116] For example, the vector of "obstetrics and gynecology" is the average of "woman emb + delivery emb + department emb".

[0117] - Word segmentation tool: Confirm the boundaries of words. For example, combine "woman", "delivery", "department" into "obstetrics and gynecology" to generate the embedding representation of the complete word.

[0118] Multi-intent classification module:

[0119] - The embedding vectors at each non-MASK position are fed into the Softmax layer for multi-classification.

[0120] - The classification result outputs labels (such as "position", "industry", "skill", etc.), supporting the simultaneous parsing of multiple intents.

[0121] - The label numbers (such as "position: 0", "industry: 2") are associated with the classification results for standardization processing.

[0122] Output layer:

[0123] -The system ultimately outputs multiple category labels (Label1, Label2, etc.) to represent different intents in the query and provide a basis for subsequent personalized recommendations.

[0124] Input example:

[0125] The user input query statement is: "obstetrician and gynecologist", and the user portrait information is: "Industry: medical, job preference: doctor".

[0126] Processing flow:

[0127] Data input module:

[0128] -The query statement and user profile information are concatenated to form an input sequence: "obstetrician and gynecologist [seq] medical doctor".

[0129] Semantic parsing module:

[0130] -Transformer Encoder performs deep semantic analysis on the input sequence and generates an embedding vector for each token:

[0131] …

[0132] gynecologyemb,obstetricsemb,scienceemb,medicalemb,birthemb,[seq]emb,medicalemb,doctoremb

[0133] …

[0134] Prefix array processing module:

[0135] -Prefix array calculation interval sum:

[0136] …

[0137] Gynecology and obstetrics emb = (gynecology emb + obstetrics emb + department of obstetrics emb) / 3

[0138] Doctor emb = (doctor emb + student emb) / 2

[0139] …

[0140] -After the word segmentation tool confirms the boundary, it generates complete word embedding representations: "obstetrics and gynecology emb" and "doctor emb".

[0141] Multi-intent classification module:

[0142] - Send the word embedding vector of each non-MASK position to the Softmax layer for multi-classification:

[0143] …

[0144] Obstetrics and Gynecology->Label:Industry

[0145] Doctor->Label:Position

[0146] …

[0147] Personalized matching module:

[0148] -Dynamically adjust weights based on user profile information (industry preference: medical; position preference: doctor):

[0149] …

[0150] Industry weighting increased-> medical-related positions are recommended first

[0151] Position weight increased->prioritized recommendation of doctor-related positions

[0152] …

[0153] Output:

[0154] The system outputs the analysis results:

[0155] …

[0156] Tag 1: Industry -> Medical

[0157] Tag 2: Position -> Doctor

[0158] …

[0159] Example of recommended results:

[0160] Based on the analysis results, eligible positions are recommended to users, such as "obstetrician and gynecologist position".

[0161] Through the above technical solution of the present invention, the system can efficiently realize the following functions:

[0162] Deep semantic understanding: Accurately capture the complex semantics in user queries.

[0163] Multi-intent separation: supports simultaneous recognition of multiple intents to improve parsing capabilities.

[0164] Personalized recommendation optimization: Dynamically adjust the matching strategy based on user portraits to improve recommendation accuracy.

[0165] Furthermore, the semantic parsing module implements deep semantic parsing through the following steps:

[0166] The user's query statement and user profile information are concatenated using the special character [seq] to form an input sequence;

[0167] Transformer Encoder is used to vectorize the input sequence and generate an embedding vector for each basic unit token. The prefix array preSum is used to calculate the sum of word vectors in any interval, including: after confirming the word boundary, the formula preSum[j]-preSum[i-1] is used to quickly calculate the word vector of the interval [i,j]. The vectors in the interval are averaged to generate an embedding representation of the complete word. Combined with context information, the dependency relationship between words in the query statement is captured through the self-attention mechanism to enhance the context perception ability of the semantic representation. Position encoding is introduced to incorporate word order information into word embedding so that the model can capture the word order structure characteristics. The multi-intent classification module realizes multi-intent recognition through the following steps: the embedding vector of each non-MAS position generated by the semantic parsing module is input into the Softmax layer for multi-classification; the Softmax layer outputs the corresponding category label, where the category label includes but is not limited to "position", "industry", "skill", "language", which is used to identify multiple intents in the query request; the identified multiple intents are separated, and an independent parsing result is generated for each intent to realize hierarchical recognition of query intent; combined with user portrait information, the weight of the classification result is dynamically adjusted to ensure that the classification result is highly relevant to user needs.

[0168] Through deep semantic analysis, multi-intent separation and personalized recommendation optimization, we provide efficient and accurate service support for recruitment platforms and job seekers. The following are the core points of the technical solution and its implementation steps:

[0169] Core technical functions

[0170] Deep semantic understanding

[0171] Input sequence construction: The user query statement and user profile information are concatenated through the special character [seq] to form a context-rich input sequence.

[0172] Transformer Encoder:

[0173] The input sequence is vectorized to generate an embedding vector for each basic unit (token).

[0174] The self-attention mechanism is introduced to capture the dependencies between words in the query statement and enhance the context perception ability.

[0175] Incorporating positional encoding enables the model to capture word order structural features.

[0176] Prefix array calculation:

[0177] By formula

[0178] The vectors in the interval are averaged to generate an embedding representation of the complete word.

[0179] Multiple intent separation

[0180] Softmax classification:

[0181] The embedding vector of each non-MASK position is input into the Softmax layer for multi-classification.

[0182] Output category labels (such as "position", "industry", "skills", etc.) and support simultaneous recognition of multiple intents.

[0183] Hierarchical recognition:

[0184] Separate the identified multiple intents and generate independent parsing results for each intent.

[0185] Dynamic weight adjustment:

[0186] Combined with user portrait information, the classification results are dynamically weighted to ensure that the classification results are highly relevant to user needs.

[0187] Personalized recommendation optimization:

[0188] Based on the analysis results, provide users with recommended content (such as positions or talents) that meet the requirements.

[0189] Dynamically adjust the recommendation strategy to prioritize content that is more relevant to user portrait information and improve recommendation accuracy.

[0190] Technical implementation steps:

[0191] Semantic parsing module:

[0192] Concatenate the query statement and user profile information to form an input sequence.

[0193] Transformer Encoder is used to generate token embedding vectors, and the embedding representation of the complete word is quickly calculated through the prefix array.

[0194] Self-attention mechanism and position encoding enhance the ability to perceive context and capture word order features.

[0195] Multi-intent classification module:

[0196] The embedded vector is input into the Softmax layer for multi-classification and the category label (such as "position", "industry", etc.) is output.

[0197] Separate multiple intents and generate independent parsing results to achieve hierarchical recognition.

[0198] Personalized matching module:

[0199] Combine user portraits and historical behavior data to dynamically adjust the weight of classification results.

[0200] Optimize recommendation strategies to provide users with accurate and highly relevant recommended content.

[0201] Technical advantages:

[0202] Efficient semantic parsing:

[0203] Based on Transformer Encoder and prefix array calculation, fast and efficient deep semantic understanding is achieved.

[0204] Multi-intent processing capabilities:

[0205] Supports separation of multiple intents in complex queries to meet diverse user needs.

[0206] Personalized recommendation optimization:

[0207] Dynamically combine user portrait information and behavioral data to improve matching accuracy and service efficiency.

[0208] Highly adaptable:

[0209] Able to quickly adapt to the query and parsing needs of different fields (such as recruitment, education, e-commerce, etc.).

[0210] Example of recommended results:

[0211] The user enters the query: "obstetrician and gynecologist", combined with the user portrait information (industry: medical; job preference: doctor).

[0212] After system analysis, the output is:

[0213] text

[0214] Tag 1: Industry -> Medical

[0215] Tag 2: Position -> Doctor

[0216] Recommendation results: Recommend eligible positions to users, such as "obstetrician and gynecologist position".

[0217] This invention solves the problems of shallow semantic understanding, weak multi-intent processing capability, and insufficient personalized matching in traditional recruitment systems through deep learning technology. The system can efficiently parse complex query requests, accurately identify multiple intents, and dynamically optimize recommendation strategies based on user portraits, providing efficient, accurate, and intelligent service experience for recruitment platforms and job seekers.

[0218] The personalized matching module implements personalized recommendations through the following steps: obtaining the user's query request and user portrait information, including but not limited to the user's industry, job preference, and skill expertise; dynamically adjusting the weights of each category based on the category labels output by the multi-intention classification module, and giving priority to matching categories with high relevance to the user portrait information; using the user's historical behavior data, including browsing history and clicking habits, to build a user interest model and further optimize the recommendation strategy; combining the adjusted category weights and user interest model to generate personalized recommendation results that meet user needs; sorting and filtering the generated personalized recommendation results, and selecting several results with the highest confidence to output to the user end. The prefix array processing module achieves efficient calculation through the following steps: constructing a prefix array preSum: preSum[j] represents the cumulative vector sum from the start of the sequence to position j; quickly calculating the cumulative sum of any interval [i, j]: using the formula preSum[j]-preSum[i-1] to calculate the cumulative sum of all token embedding vectors in the interval [i, j] in O(1) time; generating an embedding representation of a word: averaging the cumulative vectors in the interval [i, j] to obtain the embedding representation of the corresponding complete word; confirming the word boundary in conjunction with the word segmentation tool: accurately locating the start and end positions of each word through word segmentation to ensure that the interval calculation result accurately corresponds to the semantic information of the word. The system further includes a user portrait enhancement module, which is used to optimize user portrait information in the following ways: dynamically update user portraits according to the user's historical behavior data; splice user portrait information with query requests through special characters [seq] to form a more comprehensive input sequence; use user portrait information as a context reference in the semantic parsing and multi-intent classification process to improve the understanding of query intent; adjust the weights of industry, position, and skill attributes in user portraits to adapt to personalized needs in different scenarios. The system further includes a multi-label mapping module, which is used to label the intent categories in the query request, specifically including: mapping the intent categories in the query request with a predefined label set, the label set including but not limited to "position", "rank", "industry", "product", "major", "company", "school", "language", "skills"; assigning a unique label number to each intent category and associating it with the output result of the multi-intent classification module; supporting the output of query parsing results with multiple labels to meet the expression requirements of multiple intents in user queries; optimizing the processing efficiency of the subsequent personalized recommendation module through the mapping relationship between label numbers and classification results.The described system further includes a tag priority sorting module for sorting the category tags output by the multi-intent classification module, specifically including: assigning initial weights to the category tags according to user profile information; dynamically adjusting the priorities of the category tags according to the user's historical behavior data; sorting the multiple category tags output by the multi-intent classification module from high to low according to the priorities; and using the sorted priority tags as the input of the personalized recommendation module to optimize the relevance and accuracy of the recommendation results. For example. Figure 2 As shown, a query parsing system based on Transformer Encoder performs in-depth semantic parsing and multi-intent classification by combining a prefix array and a word segmentation tool. The specific meaning is as follows:

[0219] Input layer (Transformer Encoder)

[0220] The input is the query statement submitted by the user, such as "obstetrician and gynecologist".

[0221] The query is split into basic units (tokens), such as "fu", "chan", "ke", "yi", "sheng", and corresponding embedding vectors (embeddings) are generated through Transformer Encoder, such as "fu emb", "chan emb", etc.

[0222] Word segmentation tool and prefix array

[0223] The word segmentation tool is used to identify the boundaries of words. For example, "fu", "chan", "ke" are combined into "obstetrics and gynecology", and "yi", "sheng" are combined into "doctor".

[0224] The prefix array (preSum) is used to quickly calculate the interval sum of word vectors:

[0225] For example, the vector of "obstetrics and gynecology" is the average value of "fu emb + chan emb + ke emb".

[0226] Through preSum, the word vectors of any interval can be efficiently obtained, avoiding repeated calculations.

[0227] Middle layer (embedding and sequence processing)

[0228] After word segmentation and prefix array processing, the embedding representation of each complete word is generated, such as "obstetrics and gynecology emb" and "doctor emb".

[0229] The special character [seq] is used to splice the user query and user profile information (such as industry, position, etc.) to enrich the context information.

[0230] Output layer (Softmax classification)

[0231] The embedding vector of each non-MASK position is multi-classified through the Softmax layer and the corresponding category label (Label1, Label2, etc.) is output.

[0232] Tags correspond to query intents, such as "position", "skills", etc., and support simultaneous parsing of multiple intents.

[0233] User Profile Association

[0234] User portrait data (such as position, industry, skills, etc.) is spliced ​​into the query through special characters to further enhance personalized parsing capabilities.

[0235] like Figure 3 As shown in the figure, the technical method of using the prefix sum array (preSum) to calculate the interval sum of the word vectors in the query statement is shown in the figure. The specific meaning is as follows:

[0236] Definition and function of prefix and array:

[0237] -The figure marks preSum[j] and preSum[i-1], which represent the cumulative vector sum from the start of the sequence to position j and position i-1, respectively.

[0238] -By calculating the prefix sum, we can quickly get the vector sum of any interval [i,j], the formula is:

[0239] \text{Interval sum}=\text{preSum}[j]-\text{preSum}[i-1]

[0240] -This method avoids repeated calculations and improves processing efficiency.

[0241] Generation and boundary confirmation of word vectors:

[0242] -Each word in the query generates a corresponding embedding vector (token embedding) through a deep learning model (such as BERT).

[0243] -The blue part in the figure represents the previous word, and the red part represents the currently processed interval [i,j].

[0244] -The word segmentation tool is used to identify the boundaries of each word and ensure that the vectors within the interval are summed to obtain the complete word vector.

[0245] Average calculation of interval vector:

[0246] -For the interval [i,j], add up all the token vectors and take the average value to generate the word vector corresponding to the interval. For example, in the figure, the overall embedding representation of "obstetrician and gynecologist" can be calculated by the interval sum.

[0247] Multi-intent classification:

[0248] -The vector of each non-MASK position (i.e. valid position) is multi-classified through the softmax layer to identify multiple intents in the query.

[0249] -This method can effectively separate different intents in complex queries and improve parsing capabilities.

[0250] This figure shows how to quickly calculate the word vectors in a query statement through prefixes and arrays, and how to combine word segmentation tools with soft max classification to achieve efficient, multi-intent parsing. This solution can improve semantic understanding capabilities while optimizing information processing efficiency.

[0251] like Figure 4 As shown, the method for automatically parsing query requests submitted by job seekers or recruiters as described above includes the following steps: receiving query requests: receiving query requests submitted by job seekers or recruiters, and concatenating the query statements and user portrait information through special characters to form an input sequence; deep semantic analysis: using a deep learning model to perform deep semantic analysis on the input sequence, decomposing the query statements into multiple basic units, and generating corresponding embedding vectors; multi-intention classification: inputting the embedding vector generated by the semantic analysis module into the multi-intention classification module, performing multi-classification on the input, identifying multiple intents in the query request and outputting corresponding category labels; dynamic updating of user portraits: dynamically updating user portrait information based on the user's historical behavior data to reflect the user's latest needs and preferences; generating personalized recommendations: combining the updated user portrait information and the multi-intention classification results, adjusting the weights of each category, generating personalized recommendation results that meet user needs, and sorting and filtering the results, and outputting them to the user end.

[0252] The following is an interpretation of actual cases:

[0253] Case Study

[0254] Input example:

[0255] The user input query statement is: "obstetrician and gynecologist", and the user portrait information is: "Industry: medical, job preference: doctor".

[0256] Processing flow:

[0257] The data input module concatenates the query statement and the user portrait into an input sequence: "obstetrician and gynecologist [seq] medical doctor".

[0258] The semantic parsing module uses Transformer Encoder to generate an embedding vector for each token and calculates the embedding representation of the complete word through the prefix array, such as "obstetrics and gynecology emb" and "doctor emb".

[0259] The multi-intent classification module identified two category labels: “Industry: Medical” and “Position: Doctor”.

[0260] The personalized matching module dynamically adjusts the weight according to the user portrait and gives priority to recommending doctor positions in medical-related positions.

[0261] Output:

[0262] The system finally outputs the analysis results:

[0263] Tag 1: Industry -> Medical

[0264] Tag 2: Position -> Doctor

[0265] Example of recommendation result: Recommending eligible positions to users, such as "obstetrician and gynecologist position".

[0266] This method achieves efficient parsing and precise matching of complex query requests through deep semantic analysis, multi-intent classification and personalized recommendation optimization, providing intelligent service support for recruitment platforms and job seekers.

[0267] The above contents are further detailed descriptions of the present invention in combination with specific implementation methods. It cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the field of data processing technology to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the protection scope determined by the claims submitted for the present invention.

Claims

1. A system for automatically parsing query requests submitted by job seekers or recruiters, characterized in that: The data input module is used to receive the query request submitted by the job seeker or the recruiter, and concatenate the query statement and the user profile information through the special character [seq] to form an input sequence; The semantic parsing module performs deep semantic parsing on the input sequence based on the Transformer Encoder model, decomposes the query statement into several basic unit tokens, and generates corresponding embedding vectors; The prefix array processing module calculates the interval sum of word vectors through the prefix array preSum, including: Confirm word boundaries using word segmentation tools; The word vector of any interval [i,j] is quickly calculated using the formula preSum[j]-preSum[i-1], and the vectors in the interval are averaged to generate the embedding representation of the complete word; The multi-intent classification module inputs the generated embedding vector of each non-MASK position into the Softmax layer for multi-classification to identify multiple intents in the query request and output the corresponding category labels; The personalized matching module combines user portrait information and historical behavior data to dynamically adjust the weight of classification results, optimize matching strategies, and generate personalized recommendation results that meet user needs.

2. The system for automatically parsing query requests submitted by job seekers or employers according to claim 1, characterized in that: The semantic parsing module implements deep semantic parsing through the following steps: The user's query statement and user profile information are concatenated using the special character [seq] to form an input sequence; Use Transformer Encoder to vectorize the input sequence and generate an embedding vector for each basic unit token; The word vector sum of any interval is calculated through the prefix array preSum, including: After confirming the word boundary, use the formula preSum[j]-preSum[i-1] to quickly calculate the word vector of the interval [i,j]; Average the vectors in the interval to generate an embedding representation of the complete word; Combined with context information, the self-attention mechanism is used to capture the dependencies between words in the query sentence, thus enhancing the context-awareness of semantic representation. Position encoding is introduced to incorporate word order information into word embedding, so that the model can capture the word order structure characteristics.

3. The system for automatically parsing query requests submitted by job seekers or employers according to claim 1, characterized in that: The multi-intent classification module implements multi-intent recognition through the following steps: The embedding vector of each non-MAS position generated by the semantic parsing module is input into the Softmax layer for multi-classification; The Softmax layer outputs the corresponding category labels, which include but are not limited to "position", "industry", "skill", and "language", which are used to identify multiple intents in the query request; Separate the multiple identified intents and generate independent parsing results for each intent to achieve hierarchical identification of query intent; Combined with user portrait information, the weight of the classification results is dynamically adjusted to ensure that the classification results are highly relevant to user needs.

4. The system for automatically parsing query requests submitted by job seekers or employers according to claim 1, characterized in that: The personalized matching module implements personalized recommendations through the following steps: Obtain user query requests and user profile information, including but not limited to the user's industry, job preferences, and skills and expertise; According to the category labels output by the multi-intent classification module, the weights of each category are dynamically adjusted to prioritize the matching of categories with higher relevance to the user profile information; Use user historical behavior data, including browsing history and click habits, to build a user interest model and further optimize the recommendation strategy; Combine the adjusted category weights and user interest model to generate personalized recommendation results that meet user needs; The generated personalized recommendation results are sorted and filtered, and the results with the highest confidence are selected and output to the user end.

5. The system for automatically parsing query requests submitted by job seekers or employers according to claim 1, characterized in that: The prefix array processing module achieves efficient calculation by the following steps: Construct the prefix array preSum: preSum[j] represents the cumulative vector sum from the start of the sequence to position j; Quickly calculate the cumulative sum of any interval [i,j]: Use the formula preSum[j]-preSum[i-1] to calculate the cumulative sum of all token embedding vectors in the interval [i,j] in O(1) time; Generate word embedding representation: average the cumulative vectors in the interval [i, j] to get the embedding representation of the corresponding complete word; Use word segmentation tools to confirm word boundaries: Use word segmentation to accurately locate the start and end positions of each word, ensuring that the interval calculation results accurately correspond to the semantic information of the word.

6. The system for automatically parsing query requests submitted by job seekers or employers according to claim 1, characterized in that: The system further includes a user portrait enhancement module, which is used to optimize the user portrait information in the following ways: Dynamically update user profiles based on historical user behavior data; The user profile information and the query request are concatenated using the special character [seq] to form a more comprehensive input sequence; In the process of semantic analysis and multi-intent classification, user profile information is used as context reference to improve the ability to understand query intent; Adjust the weights of the industry, position, and skill attributes in the user portrait to adapt to personalized needs in different scenarios.

7. The system for automatically parsing query requests submitted by job seekers or employers according to claim 1, characterized in that: The system further includes a multi-label mapping module for labeling the intent categories in the query request, specifically including: Map the intent category in the query request with a predefined tag set, which includes but is not limited to "position", "rank", "industry", "product", "major", "company", "school", "language", and "skill"; Assign a unique label number to each intent category and associate it with the output of the multi-intent classification module; Supports the output of query parsing results with multiple labels to meet the needs of expressing multiple intentions in user queries; Through the mapping relationship between label numbers and classification results, the processing efficiency of subsequent personalized recommendation modules is optimized.

8. The system for automatically parsing query requests submitted by job seekers or employers according to claim 1, characterized in that: The system further includes a label priority sorting module, which is used to prioritize the category labels output by the multi-intent classification module, specifically including: Assign initial weights to category labels based on user profile information; Dynamically adjust the priority of category labels based on the user's historical behavior data; Sort the multiple category labels output by the multi-intent classification module from high to low priority; The sorted priority labels are used as input to the personalized recommendation module to optimize the relevance and accuracy of the recommendation results.

9. The method implemented in the system for automatically parsing a query request submitted by a job seeker or a recruiter as claimed in any one of claims 1 to 8, characterized in that: The steps include: Receiving query requests: receiving query requests submitted by job seekers or recruiters, and concatenating the query statements and user profile information through special characters to form an input sequence; Deep semantic parsing: Use deep learning models to perform deep semantic parsing on the input sequence, decompose the query statement into multiple basic units, and generate corresponding embedding vectors; Multi-intent classification: The embedding vector generated by the semantic parsing module is input into the multi-intent classification module to perform multi-classification on the input, identify multiple intents in the query request and output the corresponding category labels; Dynamic update of user portraits: Dynamically update user portrait information based on the user's historical behavior data to reflect the user's latest needs and preferences; Personalized recommendation generation: Combine the updated user portrait information and multi-intent classification results, adjust the weights of each category, generate personalized recommendation results that meet user needs, sort and filter the results, and output them to the user end.

Images