An intelligent office assistant system based on natural language processing

Abstract

The present invention discloses an intelligent office assistant system based on natural language processing, which relates to the field of natural language processing technology. When the system is in operation, it obtains user query behavior data to generate a user behavior feature vector B, and uses natural language technology to analyze the user's actual needs to generate a user intention vector I. Subsequently, a similarity calculation is performed with the entry vector R of the rules and regulations inventory to generate a matching result sim(I, R). A personalized recommendation list TC is then generated and compared with the push execution threshold TZ to accurately control the pushed content. The push execution threshold TZ is adaptively adjusted based on the user's operation record after receiving the pushed content, achieving dynamic response to changes in user needs. This provides a comprehensive and efficient solution, overcomes the problem of inaccurate matching caused by relying on keyword searches, effectively avoids the phenomenon of redundant or inconsistent pushed content, and significantly improves information matching efficiency and user satisfaction.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to an intelligent office assistant system based on natural language processing. Background Art

[0002] Chinese patent application number CN202410709352.8 discloses an efficient, AI-based power plant office system, comprising a data acquisition module, a data management module, a data analysis and processing module, a decision support module, and a data presentation module. This system integrates AI technologies such as natural language processing, machine learning, and data analysis to implement an intelligent power plant office management system. It provides intelligent data management and classification capabilities, facilitating rapid information access and sharing among employees. Leveraging machine learning and data analysis techniques, it provides accurate data support and decision-making recommendations, improving decision accuracy and efficiency. Optimizing workflows and providing intelligent assistant support reduces duplication and human error, thereby improving work efficiency and accuracy.

[0003] It can be seen that currently, most office assistant systems are based on "passive query" and users must explicitly enter questions to obtain corresponding answers. This model has significant shortcomings. First, there is a lag in information access, and users cannot actively query certain new regulations or important policies when they are not aware of them. Second, the interaction efficiency is low, and frequent input and retrieval increase the user's operating costs. Finally, existing systems have difficulty in conducting in-depth analysis and mining of user behavior data, and fail to fully utilize existing interaction records to optimize the service experience. For example, if a user searches for a specific rule or regulation multiple times, it may reflect changes in business process requirements or management blind spots. However, existing technologies are unable to capture this potential information, resulting in insufficient intelligence. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides an intelligent office assistant system based on natural language processing, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent office assistant system based on natural language processing, including a behavior data collection module, a user intention analysis module, a rule matching module, a recommendation generation module, a push strategy execution module and a feedback optimization module;

[0006] The behavior data acquisition module collects the user's query behavior data, processes and analyzes it, and obtains the user behavior feature vector B;

[0007] The user intention analysis module analyzes the user query intention using natural language technology based on the acquired user behavior feature vector B to generate a user intention vector I;

[0008] The rule matching module calculates the similarity between the user intention vector I and the item vectors R in the preset rules and regulations database to generate a matching result sim(I, R);

[0009] The recommendation generation module generates a comprehensive score Fp based on the matching result sim(I, R) and combines the user behavior feature vector B and the user intention vector I to form a personalized recommendation list TC;

[0010] The push strategy execution module extracts each comprehensive score Fp in the recommendation list TC and compares it with the preset push execution threshold TZ to determine whether the content in the recommendation list TC should be pushed to the user;

[0011] The feedback optimization module adaptively adjusts the push execution threshold TZ according to the user's operation record after the user receives the push content.

[0012] Preferably, the behavior data collection module includes a query data collection unit and a behavior feature processing unit;

[0013] The query data collection unit collects data in real time from the user's query interaction, including the query keywords input by the user, the query frequency and the query timestamp;

[0014] The query keywords are obtained by using onkeyup and onchange events to monitor the real-time input of users in the input box on the interactive front-end page. Then, natural language processing technology is used to extract key words from the text content and mark them as query keywords, which are then composed into a query keyword list {Q1, Q2, Q3, ..., Qn}.

[0015] After receiving the user's query request through the backend server, the query frequency will be recorded and stored in the database, including recording the query keyword, user ID and timestamp request data of each query. Then, the query frequency of the user's query keyword obtained in real time is queried through SQL statements, and then a query frequency list of each query keyword {F1, F2, F3, ..., Fn} is formed;

[0016] The timestamp records the time of each keyword query to form a query timestamp list of the time series {T1, T2, T3, ..., Tn}.

[0017] Preferably, the behavior feature processing unit extracts and processes the query keyword list {Q1, Q2, Q3, ..., Qn}, the query frequency list {F1, F2, F3, ..., Fn} of each query keyword, and the query timestamp list of the time series {T1, T2, T3, ..., Tn}, obtains the keyword Qu of the u-th query, the query frequency Fu of the u-th query keyword Qu, and the timestamp Tu of the u-th query keyword Qu, and performs feature processing to obtain the user behavior feature vector B of user i;

[0018] The user behavior feature vector B is obtained by the following calculation formula:

[0019] ;

[0020] Where Bi represents the user behavior feature vector B of user i, n represents the total number of user queries, log represents the logarithmic function, e represents the exponential decay function, and △Tu represents the time interval of the query timestamp Tu.

[0021] Preferably, the user intention analysis module includes an intention classification modeling unit and an intention vector generation unit;

[0022] The intention classification modeling unit defines a vectorized model by using the BGE model and the M3E model, and trains the vectorized model. The user behavior feature vector B is input into the trained vectorized model to obtain the intention category matrix E, wherein the dimension of the intention category matrix E is specifically C*d dimension, C represents the number of intent categories, and d represents the dimension of each category feature vector; the intention category matrix E is specifically ;

[0023] The number of intent categories C and the dimension d of each category feature vector are obtained by extracting the company's rules and regulations text.

[0024] Preferably, the intention vector generation unit uses the intention category matrix E and combines the user behavior feature vector B of user i to generate the intention vector I of user i through vector calculation, reflecting the intention probability distribution of the current query of user i;

[0025] The intention vector I of user i is specifically obtained by projecting the user behavior feature vector B of user i onto the intention category matrix E obtained by pre-training, calculating the correlation of the intention category C, obtaining the correlation score z matrix, and then transforming the correlation score z matrix using the Softmax normalized exponential function to obtain the intention vector I matrix of user i with C*d dimensions;

[0026] The relevance score z is obtained by the following calculation formula:

[0027] ;

[0028] Where zq represents the relevance score of the qth intent category in intent category C, represents a nonlinear activation function, including Sigmoid and ReLU, d represents the dimension of each category feature vector, k represents the k-th feature in the dimension d of each category feature vector, wk represents the global weight value of the k-th feature, eqk represents the weight value of the k-th feature in the q-th intent category in the intent category C in the intent category matrix E, Bi,k represents the k-th feature in the user behavior feature vector B of user i, and bq represents the bias term of the q-th intent category in the intent category C.

[0029] Preferably, the rule matching module includes an entry vector generating unit and a similarity calculating unit;

[0030] The item vector generating unit converts each item in the rules and regulations database into a characterized vector form by using natural language processing technology to obtain each item vector R;

[0031] The item vectors R are specifically R={R1, R2, R3, ..., Rr}.

[0032] Preferably, the similarity calculation unit calculates the matching results by obtaining all the intent categories in the user intent vector I of the user i and all the vectors in the item vector R, obtains the similarity score sim(i, p, q) between the qth intent category in the intent category C in the intent vector of the user i and the pth item vector in the item vector R, and then performs aggregation processing to generate a matching result sim(I, R);

[0033] The similarity score sim(i, p, q) is obtained by the following calculation formula:

[0034] ;

[0035] Where sim(i, p, q) represents the similarity score between the qth intent category in the intent category C in the intent vector of user i and the pth item vector in the item vector R, Ii,q represents the component of the qth intent category in the intent vector of user i, and q represents the correlation score between the pth item vector in the item vector R and the qth intent category.

[0036] The matching result sim(I, R) is obtained by The calculation formula is obtained, where sim(Ii, Rp) represents the matching result of user i on the p-th entry vector in the entry vector R, aq represents the preset weight value of the q-th intent category in the intent category C, and q represents the q-th intent category in the intent category C.

[0037] Preferably, the recommendation generation module includes a personalized recommendation generation unit;

[0038] The personalized recommendation generation unit obtains the comprehensive score Fp of the p-th item vector in each item vector R based on the matching result sim(I, R) and combines the user behavior feature vector B and the user intention vector I, and generates a personalized recommendation list TC according to the comprehensive score Fp;

[0039] The comprehensive score Fp is obtained by the following calculation formula:

[0040] ;

[0041] Where f1, f2, and f3 represent the weight of the matching result, the weight of the correlation of the user behavior feature vector, and the weight of the matching degree of the user intention vector, respectively. f1+f2+f3=1, and the specific value is set by the user. sim represents the similarity function, f represents the behavior correlation function, Bi and Bp represent the user behavior feature vectors of user i and user p, respectively, g represents the deep matching function, which is specifically used to amplify the matching degree between user intention and items in key categories and highlight highly correlated categories. Ii represents the user intention vector of user i, and Rp represents the p-th item vector in the item vector R.

[0042] The recommendation list TC is associated and marked by the comprehensive score Fp of each item vector in the item vector R, and then bubble sorted to reorganize into a recommendation list TC={(R1, F1), (R2, F2), (R3, F3), ..., (Rr, Fp)}.

[0043] Preferably, the method of comparing the recommendation list TC with the preset push execution threshold TZ is as follows:

[0044] When the comprehensive score Fp ≥ the push execution threshold TZ, the comprehensive score Fp and the associated item vectors Rr in the recommendation list TC are obtained and pushed to the user;

[0045] When the comprehensive score Fp is less than the push execution threshold TZ, the comprehensive score Fp and the associated item vectors Rr in the recommendation list TC are obtained and not pushed to the user.

[0046] Preferably, the user's operation record is obtained by extracting front-end log records to obtain the user's non-interactive behavior time period information on the recommended content, including the pop-up time TCSJ and the closing time GB, and the stay time TL is calculated. The push execution threshold TZ is proportionally adjusted according to the calculated stay time TL to obtain the secondary adjusted push execution threshold ETZ;

[0047] The length of stay TL is Obtain calculation formula;

[0048] The secondary adjustment push execution threshold ETZ is achieved by , where β represents the learning rate, and Tmax represents the preset maximum stay time of users using recommended content.

[0049] The present invention provides an intelligent office assistant system based on natural language processing, which has the following beneficial effects:

[0050] (1) When the system is running, it obtains the user's query behavior data to generate a user behavior feature vector B, and uses natural language technology to analyze the user's real needs and generate a user intention vector I. Subsequently, it calculates the similarity with the entry vector R of the rules and regulations inventory to generate a matching result sim(I, R), and then generates a personalized recommendation list TC. It compares it with the push execution threshold TZ to accurately control the push content. It uses the user's operation record after receiving the push content to adaptively adjust the push execution threshold TZ to achieve dynamic response to changes in user needs, providing a comprehensive and efficient solution. On the one hand, it uses natural language technology to analyze user needs and generate user intentions, overcoming the problem of inaccurate matching caused by relying on keyword search; on the other hand, it combines user behavior characteristics and push execution threshold TZ for recommendation and dynamic optimization, effectively avoiding the phenomenon of redundant or inconsistent push content. It achieves accurate push and personalized recommendation of rules and regulations information, greatly improving information matching efficiency and user satisfaction.

[0051] (2) By using natural language processing technology to convert the items in the rules and regulations database into item vectors R, and by matching the user intent vector I with the item vector R, the similarity score sim(i, p, C) is calculated by category, achieving a deep match between user behavior and rules and regulations items, and fully exploring the semantic relevance of rules and regulations items under multi-category features. By combining the user intent vector I with the intent category matrix E, it can provide a multi-dimensional interpretation of complex user query behavior, and through the sub-item aggregation strategy, dynamically weigh the importance of different intent categories to ensure that the matching result sim(I, R) is more in line with the user's current actual needs. Compared with the previous single-dimensional matching method, this system greatly improves the matching accuracy and relevance, optimizes the query efficiency of rules and regulations and optimizes the user experience.

[0052] (3) Based on the matching result sim(I, R), combined with the user behavior feature vector B and the user intention vector I, and by associating the scoring results with the items, a personalized recommendation list TC is generated through bubble sorting to further screen the items that meet the user's needs. For the user's non-interactive behavior time period, the push execution threshold TZ is proportionally adjusted by calculating the ratio of the stay time TL to the preset maximum stay time Tmax, generating a secondary adjustment push execution threshold ETZ, and adaptively optimizing the push strategy. Compared with the traditional static scoring and fixed push rule method, it reflects user needs at a deeper level. By adjusting the push execution threshold ETZ twice, it can adapt to the dynamic changes of user interests in real time, ensuring that the recommended content is neither redundant nor omitted, further optimizing the push effect and user satisfaction, and comprehensively improving the intelligence level of the recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a schematic block diagram of an intelligent office assistant system based on natural language processing according to the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Example

[0055] The present invention provides an intelligent office assistant system based on natural language processing, please refer to Figure 1 , including behavioral data collection module, user intention analysis module, rule matching module, recommendation generation module, push strategy execution module and feedback optimization module;

[0056] The behavior data acquisition module collects the user's query behavior data, processes and analyzes it, and obtains the user behavior feature vector B;

[0057] The user intention analysis module analyzes the user query intention using natural language technology based on the acquired user behavior feature vector B to generate a user intention vector I;

[0058] The rule matching module calculates the similarity between the user intention vector I and the item vectors R in the preset rules and regulations database to generate a matching result sim(I, R);

[0059] The recommendation generation module generates a comprehensive score Fp based on the matching result sim(I, R) and combines the user behavior feature vector B and the user intention vector I to form a personalized recommendation list TC;

[0060] The push strategy execution module extracts each comprehensive score Fp in the recommendation list TC and compares it with the preset push execution threshold TZ to determine whether the content in the recommendation list TC should be pushed to the user;

[0061] The feedback optimization module adaptively adjusts the push execution threshold TZ according to the user's operation record after the user receives the push content.

[0062] In this embodiment, the user's query behavior data is obtained to generate a user behavior feature vector B, and the user's real needs are analyzed using natural language technology to generate a user intention vector I. Subsequently, a similarity calculation is performed with the entry vector R of the rules and regulations inventory to generate a matching result sim(I, R), and then a personalized recommendation list TC is generated. This is compared with the push execution threshold TZ to accurately control the push content. The user's operation record after receiving the push content is used to adaptively adjust the push execution threshold TZ to achieve a dynamic response to changes in user needs, providing a comprehensive and efficient solution. On the one hand, natural language technology is used to analyze user needs and generate user intentions, overcoming the problem of inaccurate matching caused by relying on keyword searches; on the other hand, the combination of user behavior characteristics and push execution threshold TZ for recommendation and dynamic optimization effectively avoids the redundancy of pushed content or the phenomenon of inconsistency with needs. This achieves accurate push and personalized recommendation of rules and regulations information, greatly improving information matching efficiency and user satisfaction. Example

[0063] This embodiment is explained in Example 1, please refer to Figure 1 ,Specifically: the behavior data acquisition module includes a query data acquisition unit and a behavior feature processing unit;

[0064] The query data collection unit collects data in real time from the user's query interaction, including the query keywords input by the user, the query frequency and the query timestamp;

[0065] The query keywords are obtained by using onkeyup and onchange events to monitor the real-time input of users in the input box on the interactive front-end page. Then, natural language processing technology is used to extract key words from the text content and mark them as query keywords, which are then composed into a query keyword list {Q1, Q2, Q3, ..., Qn}.

[0066] After receiving the user's query request through the backend server, the query frequency will be recorded and stored in the database, including recording the query keyword, user ID and timestamp request data of each query. Then, the query frequency of the user's query keyword obtained in real time is queried through SQL statements, and then a query frequency list of each query keyword {F1, F2, F3, ..., Fn} is formed;

[0067] The timestamp records the time of each keyword query to form a query timestamp list of the time series {T1, T2, T3, ..., Tn}.

[0068] The behavior feature processing unit extracts and processes the query keyword list {Q1, Q2, Q3, ..., Qn}, the query frequency list {F1, F2, F3, ..., Fn} of each query keyword, and the query timestamp list of the time series {T1, T2, T3, ..., Tn}, obtains the keyword Qu of the u-th query, the query frequency Fu of the u-th query keyword Qu, and the timestamp Tu of the u-th query keyword Qu, and performs feature processing to obtain the user behavior feature vector B of user i;

[0069] The user behavior feature vector B is obtained by the following calculation formula:

[0070] ;

[0071] Where Bi represents the user behavior feature vector B of user i, which is specifically used to describe the query behavior pattern of user i, including keyword preference, query frequency, and time interval characteristics. n represents the total number of user queries, specifically the total number in the query keyword list {Q1, Q2, Q3, …, Qn}. log represents the logarithmic function, e represents the exponential decay function, and △Tu represents the time interval of the query timestamp Tu.

[0072] In this embodiment, the text content entered by the user is obtained in real time by utilizing the event monitoring technology of the interactive front-end page, and the query keywords are extracted and generated in combination with natural language processing technology. At the same time, the back-end server records the request data and timestamp information of the user's query, and the query frequency list and query timestamp list of each query keyword are counted in real time to form a high-precision behavior data record. Further, based on the combination of query keywords, query frequency and query timestamp, the user's behavior pattern is feature extracted. Finally, the user behavior feature vector B of user i is generated through formula calculation, which effectively quantifies the user's query behavior characteristics, ensures the real-time and integrity of the user query behavior data, and realizes the pattern analysis of user query behavior, making up for the lack of traditional query systems in capturing dynamic changes in behavior. By generating the user behavior feature vector B, the module can accurately describe the user's keyword preferences, query frequency changes and time interval characteristics, thereby providing richer input data. Compared with the traditional method that relies solely on keyword matching, incorporating dynamic behavior features into the user profile significantly enhances the system's ability to understand user needs globally. This transition from a single query behavior to a comprehensive behavior pattern lays a more scientific foundation for subsequent intent analysis and personalized recommendations. Example

[0073] This embodiment is explained in Example 2, please refer to Figure 1 ,Specifically: the user intention analysis module includes an intention classification modeling unit and an intention vector generation unit;

[0074] The intention classification modeling unit defines a vectorized model by using the BGE model and the M3E model, and trains the vectorized model. The user behavior feature vector B is input into the trained vectorized model to obtain the intention category matrix E, wherein the dimension of the intention category matrix E is specifically C*d dimension, C represents the number of intent categories, and d represents the dimension of each category feature vector; the intention category matrix E is specifically ;

[0075] The number of intent categories C and the dimension d of each category feature vector are extracted from the company's rules and regulations text.

[0076] The intention vector generation unit uses the intention category matrix E and the user behavior feature vector B of user i to generate the intention vector I of user i through vector calculation, reflecting the intention probability distribution of the current query of user i;

[0077] The intention vector I of user i is specifically obtained by projecting the user behavior feature vector B of user i onto the intention category matrix E obtained by pre-training, calculating the correlation of the intention category C, obtaining the correlation score z matrix, and then transforming the correlation score z matrix using the Softmax normalized exponential function to obtain the intention vector I matrix of user i with C*d dimensions;

[0078] The relevance score z is obtained by the following calculation formula:

[0079] ;

[0080] Where zq represents the relevance score of the qth intent category in intent category C, represents a nonlinear activation function, including Sigmoid and ReLU, d represents the dimension of each category feature vector, k represents the k-th feature in dimension d of each category feature vector, wk represents the global weight value of the k-th feature, eqk represents the weight value of the k-th feature in the q-th intent category in intent category C in the intent category matrix E, Bi,k represents the k-th feature in the user behavior feature vector B of user i, and bq represents the bias term of the q-th intent category in intent category C;

[0081] The relevance score zC of the intent category C is specifically ;

[0082] By integrating the relevance scores zC of the intent category C, a relevance score z matrix of the C*d dimension is obtained. The relevance score z matrix is ​​specifically: ;

[0083] The intention vector I matrix of user i is specifically: .

[0084] The rule matching module includes an entry vector generating unit and a similarity calculating unit;

[0085] The item vector generating unit converts each item in the rules and regulations database into a characterized vector form by using natural language processing technology to obtain each item vector R;

[0086] The item vectors R are specifically R={R1, R2, R3, ..., Rr}.

[0087] The similarity calculation unit calculates the matching results by obtaining all the intent categories in the user intention vector I of user i and all the vectors in the item vector R, obtains the similarity score sim(i, p, q) between the qth intent category in the intent category C in the intention vector of user i and the pth item vector in the item vector R, and then performs aggregation processing to generate a matching result sim(I, R);

[0088] The similarity score sim(i, p, q) is obtained by the following calculation formula:

[0089] ;

[0090] Where sim(i, p, q) represents the similarity score between the qth intent category in the intent category C in the intent vector of user i and the pth item vector in the item vector R, Ii,q represents the component of the qth intent category in the intent vector of user i, and Rp,q represents the correlation score between the pth item vector in the item vector R and the qth intent category.

[0091] The matching result sim(I, R) is obtained by The calculation formula is obtained, where sim(Ii, Rp) represents the matching result of user i on the p-th item vector in the item vector R, aq represents the preset weight value of the q-th intent category in the intent category C, and q represents the q-th intent category in the intent category C;

[0092] The matching result sim(I, R) is specifically sim(I, R)={sim(Ii, R1), sim(Ii, R2), sim(Ii, R3), ..., sim(Ii, Rr)}.

[0093] This embodiment comprehensively improves the relevance of rules and regulations to user needs through precise intent parsing and efficient item matching. Combining user i's user behavior feature vector B with a pre-trained intent category matrix E, the intent vector I for user i is generated through nonlinear mapping and a Softmax normalized exponential function, transforming the user's behavior pattern into a multi-category intent probability distribution. Simultaneously, natural language processing techniques are used to convert the items in the rules and regulations database into item vectors R. By matching the user intent vector I with the item vector R, a category-by-category similarity score sim(i, p, C) is calculated. This achieves a deep match between user behavior and rule and regulation items, fully exploring the semantic relevance of rule and regulation items under multi-category features. By combining the user intent vector I with the intent category matrix E, a multi-dimensional interpretation of complex user query behavior is provided. Through a sub-item aggregation strategy, the importance of different intent categories is dynamically weighed, ensuring that the matching result sim(I, R) better meets the user's current actual needs. Compared to previous single-dimensional matching methods, this system significantly improves matching accuracy and relevance, optimizing rule and regulation query efficiency and user experience. Example

[0094] This embodiment is explained in Example 3, please refer to Figure 1 ,Specifically: the recommendation generation module includes a personalized recommendation generation unit;

[0095] The personalized recommendation generation unit obtains the comprehensive score Fp of the p-th item vector in each item vector R based on the matching result sim(I, R) and combines the user behavior feature vector B and the user intention vector I, and generates a personalized recommendation list TC according to the comprehensive score Fp;

[0096] The comprehensive score Fp is obtained by the following calculation formula:

[0097] ;

[0098] Wherein, f1, f2 and f3 represent the weight value of the matching result, the weight value of the correlation of the user behavior feature vector and the weight value of the matching degree of the user intention vector, respectively, and f1+f2+f3=1. The specific value is set by the user. sim represents the similarity function, which is specifically used to provide a global matching score to measure the overall fit between user intention and item content. f represents the behavior correlation function, which is specifically used to reflect the matching degree between user historical behavior and item content and strengthen the behavioral basis of recommendation. Bi and Bp represent the user behavior feature vectors of user i and user p, respectively. g represents the deep matching function, which is specifically used to amplify the matching degree between user intention and item in key categories and highlight highly correlated categories. Ii represents the user intention vector of user i. Rp represents the p-th item vector in the item vector R.

[0099] The recommendation list TC is associated and marked by the comprehensive score Fp of each item vector in the item vector R, and then bubble sorted to reorganize into a recommendation list TC={(R1, F1), (R2, F2), (R3, F3), ..., (Rr, Fp)}.

[0100] The method of comparing the recommendation list TC with the preset push execution threshold TZ is as follows:

[0101] When the comprehensive score Fp ≥ the push execution threshold TZ, the comprehensive score Fp and the associated item vectors Rr in the recommendation list TC are obtained and pushed to the user;

[0102] When the comprehensive score Fp is less than the push execution threshold TZ, the comprehensive score Fp and the associated item vectors Rr in the recommendation list TC are obtained and not pushed to the user.

[0103] The user's operation record is extracted from the front-end log record to obtain the user's non-interactive behavior time period information for the recommended content, including the pop-up time TCSJ and the closing time GB, and the stay time TL is calculated. The push execution threshold TZ is proportionally adjusted according to the calculated stay time TL to obtain the secondary adjusted push execution threshold ETZ;

[0104] The length of stay TL is Obtain calculation formula;

[0105] The secondary adjustment push execution threshold ETZ is achieved by , where β represents the learning rate, which is used to control the amplitude of the proportional adjustment, and Tmax represents the preset maximum stay time of users using the recommended content.

[0106] In this embodiment, based on the matching result sim(I, R), combined with the user behavior feature vector B and the user intent vector I, the scoring results are associated with the items. A bubble sort is used to generate a personalized recommendation list TC to further filter items that meet user needs. For the user's non-interactive behavior time period, the push execution threshold TZ is proportionally adjusted by calculating the ratio of the dwell time TL to the preset maximum dwell time Tmax. This generates a secondary adjustment push execution threshold ETZ, adaptively optimizing the push strategy. Compared with traditional static scoring and fixed push rule methods, this method reflects user needs at a deeper level. By secondary adjustment of the push execution threshold ETZ, it can adapt to the dynamic changes of user interests in real time, ensuring that recommended content is neither redundant nor omitted, further optimizing push effectiveness and user satisfaction, and comprehensively improving the intelligence level of the recommendation system.

[0107] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent office assistant system based on natural language processing, characterized by: It includes behavioral data collection module, user intention analysis module, rule matching module, recommendation generation module, push strategy execution module and feedback optimization module; The behavior data acquisition module collects the user's query behavior data, processes and analyzes it, and obtains the user behavior feature vector B; The user intention analysis module analyzes the user query intention using natural language technology based on the acquired user behavior feature vector B to generate a user intention vector I; The rule matching module calculates the similarity between the user intention vector I and the item vectors R in the preset rules and regulations database to generate a matching result sim(I, R); The recommendation generation module generates a comprehensive score Fp based on the matching result sim(I, R) and combines the user behavior feature vector B and the user intention vector I to form a personalized recommendation list TC; The push strategy execution module extracts each comprehensive score Fp in the recommendation list TC and compares it with the preset push execution threshold TZ to determine whether the content in the recommendation list TC should be pushed to the user; The feedback optimization module adaptively adjusts the push execution threshold TZ according to the user's operation record after the user receives the push content.

2. The intelligent office assistant system based on natural language processing according to claim 1, characterized in that: The behavior data collection module includes a query data collection unit and a behavior feature processing unit; The query data collection unit collects data in real time from the user's query interaction, including the query keywords input by the user, the query frequency and the query timestamp; The query keywords are obtained by using onkeyup and onchange events to monitor the real-time input of users in the input box on the interactive front-end page. Then, natural language processing technology is used to extract key words from the text content and mark them as query keywords, which are then composed into a query keyword list {Q1, Q2, Q3, ..., Qn}. After receiving the user's query request through the backend server, the query frequency will be recorded and stored in the database, including recording the query keyword, user ID and timestamp request data of each query. Then, the query frequency of the user's query keyword obtained in real time is queried through SQL statements, and then a query frequency list of each query keyword {F1, F2, F3, ..., Fn} is formed; The timestamp records the time of each keyword query to form a query timestamp list of the time series {T1, T2, T3, ..., Tn}.

3. The intelligent office assistant system based on natural language processing according to claim 2, characterized in that: The behavior feature processing unit extracts and processes the query keyword list {Q1, Q2, Q3, ..., Qn}, the query frequency list {F1, F2, F3, ..., Fn} of each query keyword, and the query timestamp list of the time series {T1, T2, T3, ..., Tn}, obtains the keyword Qu of the u-th query, the query frequency Fu of the u-th query keyword Qu, and the timestamp Tu of the u-th query keyword Qu, and performs feature processing to obtain the user behavior feature vector B of user i; The user behavior feature vector B is obtained by the following calculation formula: ； Where Bi represents the user behavior feature vector B of user i, n represents the total number of user queries, log represents the logarithmic function, e represents the exponential decay function, and △Tu represents the time interval of the query timestamp Tu.

4. The intelligent office assistant system based on natural language processing according to claim 3, characterized in that: The user intention analysis module includes an intention classification modeling unit and an intention vector generation unit; The intent classification modeling unit defines a vectorized model by using the BGE model and the M3E model, trains the vectorized model, inputs the user behavior feature vector B into the trained vectorized model, and obtains the intent category matrix E, where the dimension of the intent category matrix E is specifically C*d, where C represents the number of intent categories and d represents the dimension of the feature vector of each category; The intention category matrix E is specifically: ; The number of intent categories C and the dimension d of each category feature vector are extracted from the company's rules and regulations text.

5. The intelligent office assistant system based on natural language processing according to claim 4, characterized in that: The intention vector generation unit uses the intention category matrix E and the user behavior feature vector B of user i to generate the intention vector I of user i through vector calculation, reflecting the intention probability distribution of the current query of user i; The intention vector I of user i is specifically obtained by projecting the user behavior feature vector B of user i onto the intention category matrix E obtained by pre-training, calculating the correlation of the qth intention category in the intention category C, obtaining the correlation score z matrix, and then transforming the correlation score z matrix using the Softmax normalized exponential function to obtain the C*d dimension user i intention vector I matrix; The relevance score z is obtained by the following calculation formula: ； Where zq represents the relevance score of the qth intent category in intent category C, represents a nonlinear activation function, including Sigmoid and ReLU, d represents the dimension of each category feature vector, k represents the k-th feature in the dimension d of each category feature vector, wk represents the global weight value of the k-th feature, eqk represents the weight value of the k-th feature in the q-th intent category in the intent category C in the intent category matrix E, Bi,k represents the k-th feature in the user behavior feature vector B of user i, and bq represents the bias term of the q-th intent category in the intent category C.

6. The intelligent office assistant system based on natural language processing according to claim 5, characterized in that: The rule matching module includes an entry vector generating unit and a similarity calculating unit; The item vector generating unit converts each item in the rules and regulations database into a characterized vector form by using natural language processing technology to obtain each item vector R; The item vectors R are specifically R={R1, R2, R3, ..., Rr}.

7. The intelligent office assistant system based on natural language processing according to claim 6, characterized in that: The similarity calculation unit calculates the matching results by obtaining all the intent categories in the user intention vector I of user i and all the vectors in the item vector R, obtains the similarity score sim(i, p, q) between the qth intent category in the intent category C in the intention vector of user i and the pth item vector in the item vector R, and then performs aggregation processing to generate a matching result sim(I, R); The similarity score sim(i, p, q) is obtained by the following calculation formula: ； Where sim(i, p, q) represents the similarity score between the qth intent category in the intent category C in the intent vector of user i and the pth item vector in the item vector R, Ii,q represents the component of the qth intent category in the intent vector of user i, and Rp,q represents the correlation score between the pth item vector in the item vector R and the qth intent category. The matching result sim(I, R) is obtained by The calculation formula is obtained, where sim(Ii, Rp) represents the matching result of user i on the p-th entry vector in the entry vector R, aq represents the preset weight value of the q-th intent category in the intent category C, and q represents the q-th intent category in the intent category C.

8. The intelligent office assistant system based on natural language processing according to claim 7, characterized in that: The recommendation generation module includes a personalized recommendation generation unit; The personalized recommendation generation unit obtains the comprehensive score Fp of the p-th item vector in each item vector R based on the matching result sim(I, R) and combines the user behavior feature vector B and the user intention vector I, and generates a personalized recommendation list TC according to the comprehensive score Fp; The comprehensive score Fp is obtained by the following calculation formula: ； Where f1, f2, and f3 represent the weight of the matching result, the weight of the correlation of the user behavior feature vector, and the weight of the matching degree of the user intention vector, respectively. f1+f2+f3=1, and the specific value is set by the user. sim represents the similarity function, f represents the behavior correlation function, Bi and Bp represent the user behavior feature vectors of user i and user p, respectively, g represents the deep matching function, which is specifically used to amplify the matching degree between user intention and items in key categories and highlight highly correlated categories. Ii represents the user intention vector of user i, and Rp represents the p-th item vector in the item vector R. The recommendation list TC is associated and marked by the comprehensive score Fp of each item vector in the item vector R, and then bubble sorted to reorganize into a recommendation list TC={(R1, F1), (R2, F2), (R3, F3), ..., (Rr, Fp)}.

9. The intelligent office assistant system based on natural language processing according to claim 8, characterized in that: The method of comparing the recommendation list TC with the preset push execution threshold TZ is as follows: When the comprehensive score Fp ≥ the push execution threshold TZ, the comprehensive score Fp and the associated item vectors Rr in the recommendation list TC are obtained and pushed to the user; When the comprehensive score Fp is less than the push execution threshold TZ, the comprehensive score Fp and the associated item vectors Rr in the recommendation list TC are obtained and not pushed to the user.

10. The intelligent office assistant system based on natural language processing according to claim 1, characterized in that: The user's operation record is extracted from the front-end log record to obtain the user's non-interactive behavior time period information for the recommended content, including the pop-up time TCSJ and the closing time GB, and the stay time TL is calculated. The push execution threshold TZ is proportionally adjusted according to the calculated stay time TL to obtain the secondary adjusted push execution threshold ETZ; The length of stay TL is Obtain calculation formula; The secondary adjustment push execution threshold ETZ is achieved by , where β represents the learning rate, and Tmax represents the preset maximum stay time of users using recommended content.

Images