Employee search recommendation content generation method, device and equipment
By combining employee work data and historical behavior, enterprise planning and industry data, using search intention prediction models to generate recommended content, the problem of insufficient search intention prediction in work scenarios in the existing technology is solved, and more accurate and comprehensive search recommendations are achieved , improves employees' work efficiency and experience.
Patent Information
- Application Number
- CN202411496867.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-10-25
AI Technical Summary
The existing technology cannot accurately analyze user search intentions in work scenarios, resulting in the recommended content not meeting the actual needs of employees and lacking targetedness and comprehensiveness.
By obtaining employee work data and historical behaviors, combining the company's business planning and industry data, we use search intent prediction models to generate recommended content, and dynamically adjust to meet employees' current search needs.
It improves the pertinence, comprehensiveness and accuracy of search recommendations, ensures that employees can obtain the latest news from relevant industries or enterprises in a timely manner, and improves work efficiency and experience.
Smart Images

Figure CN119003891A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of recommendation technology, and in particular to a method, device and equipment for generating employee search recommendation content. Background Art
[0002] Although the existing technology can predict the user's desired search content based on the keywords entered by the user in the input box or the historical search records; however, the search recommendation considerations based only on the historical search records or the input keywords are relatively simple and cannot accurately analyze the user's search intention; especially when the existing search prediction or intention prediction methods are applied in work scenarios, accurate results are often not obtained, which leads to the recommendation content based on the prediction results being not the content that the employees are really interested in. Summary of the invention
[0003] The purpose of the embodiments of the present application is to provide a method, device and equipment for generating employee search recommendation content, so as to solve the above-mentioned problems existing in the prior art and to accurately and comprehensively generate employee search recommendation content.
[0004] In a first aspect, a method for generating employee search recommendation content is provided, and the method may include: A method for generating employee search recommendation content, characterized in that the method comprises: Obtaining the work data of the target employee and the historical behaviors performed on the work computer during the target time period, as well as the business planning data of the enterprise where the target employee works and the industry data of the industry where the enterprise works during the target time period; Analyze the historical behavior to obtain historical behavior analysis results; Inputting the historical behavior analysis results and the work data into a trained search intention prediction model to obtain a search intention prediction result of a target employee; Based on the search intention prediction results, the business planning data and the industry data, search recommendation content for target employees is generated.
[0005] In an optional implementation, the historical behavior is analyzed to obtain historical behavior analysis results, including: For any historical behavior, count the number of times the historical behavior occurs in the target time period; If the number of times is greater than the configured number threshold, the historical behavior is used as the target historical behavior; Extracting target historical associated behaviors associated with the target historical behaviors from all historical behaviors of the target employee within the target time period; Performing correlation analysis on the target historical behavior and the corresponding target historical associated behavior to obtain a correlation coefficient between the target historical associated behavior and the corresponding target historical behavior; Different target historical behaviors and corresponding target historical associated behaviors and association coefficients are input into the trained behavior analysis neural network to obtain the historical behavior analysis results.
[0006] In an optional implementation, any historical behavior corresponds to a point in time; From all historical behaviors of the target employee within the target time period, target historical associated behaviors associated with the target historical behaviors are extracted, including: Extracting the time point of the target historical behavior; Add or subtract the configured time threshold from the time point of the target historical behavior to obtain the associated time period; Performing correlation analysis on all historical behaviors of the target employee in the associated time period and the target historical behavior to obtain initial associated behaviors of the target historical behavior; Calculate the comprehensive association score between each initial association behavior and the target historical behavior; The target historical association behaviors associated with the target historical behaviors are screened out from the initial association behaviors according to the comprehensive association score and the configured comprehensive association score screening threshold.
[0007] In an optional implementation, before inputting the historical behavior analysis results and the work data into the trained search intent prediction model, the method further includes: Convert the historical behavior analysis result and the work data respectively to obtain a behavior feature vector and a work feature vector; Based on the behavior feature vector and the work feature vector, a dynamic vector of the target employee is generated.
[0008] In an optional implementation, the historical behavior analysis result and the work data are input into a trained search intent prediction model, including: The dynamic vector is input into the trained search intention prediction model to obtain the search intention prediction result of the target employee.
[0009] In an optional implementation, generating search recommendation content for target employees based on the search intention prediction result, the business planning data, and the industry data includes: Convert the business planning data and the industry data respectively to obtain an enterprise feature vector and an industry feature vector; Calculating the similarity between the dynamic vector and the enterprise feature vector and the industry feature vector respectively; Based on the similarity calculation results, generate initial search recommendation content; Based on the search intention prediction result, the initial search recommendation content is adjusted to obtain the search recommendation content for the target employee.
[0010] In a third aspect, a device for generating employee search recommendation content is provided, the device comprising: An acquisition unit, used to acquire work data of a target employee and historical behaviors performed on a work computer during a target time period, as well as business planning data of the enterprise where the target employee works and industry data of the industry where the enterprise works during the target time period; An analysis unit, used to analyze the historical behavior and obtain a historical behavior analysis result; A prediction unit, used to input the historical behavior analysis result and the work data into a trained search intention prediction model to obtain a search intention prediction result of a target employee; A generating unit is used to generate search recommendation content for target employees based on the search intention prediction result, the business planning data and the industry data.
[0011] In a third aspect, an electronic device is provided, the electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; The processor is used to implement any method step described in the first aspect when executing the program stored in the memory.
[0012] In a fourth aspect, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, any method step described in the first aspect is implemented.
[0013] The method for generating employee search recommendation content of the present application fills the gap in the prior art that there is no search intention prediction and recommendation method for work scenarios; it predicts the search intention of employees in a comprehensive and comprehensive manner through their work data, historical behavior, and the company's planning and market, and generates recommended content based on the search intention prediction results, which not only includes content that employees expect to see, but also includes important content that is closely related to the company and industry where the employees are located; it not only improves the pertinence, comprehensiveness and accuracy of search recommendations, but also can effectively improve employees' work efficiency, improve employees' experience, and enable employees to obtain the latest news in the industry or company in a timely manner.
[0014] This application generates dynamic vectors based on employees' historical behaviors and work data, taking into account various aspects of employees' information, and can more comprehensively reflect employees' needs and backgrounds; and the dynamic vectors are relatively stable and do not change frequently, thereby improving the stability and comprehensiveness of recommended content; and combining the search intent vector with the relevant content of the enterprise and industry can more timely reflect employees' current search needs, and can dynamically adjust recommended content according to different search scenarios, which is more flexible; and using search intent to adjust recommended content improves the relevance of recommendations. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0016] Figure 1 A flowchart of a method for generating employee search recommendation content provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a device for generating employee search recommendation content provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0017] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0018] The method for generating employee search recommendation content provided in the embodiment of the present application can be applied in a server or in a terminal with strong computing power. The server can be a physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms. The terminal can be a user equipment (User Equipment, UE) such as a mobile phone, a smart phone, a laptop, a digital broadcast receiver, a personal digital assistant (PDA), a tablet computer (PAD), a handheld device, a vehicle-mounted device, a wearable device, a computing device or other processing equipment connected to a wireless modem, a mobile station (Mobile Station, MS), a mobile terminal (Mobile Terminal), etc. The terminal and the server can be directly or indirectly connected through a wired or wireless communication method, which is not limited in this application.
[0019] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.
[0020] Figure 1 The flowchart of the method for generating employee search recommendation content provided in the embodiment of the present application is as follows. Figure 1 As shown, the method may include: Step S110, obtaining the work data of the target employee and the historical behaviors performed on the work computer during the target time period, as well as the business planning data of the enterprise where the target employee works and the industry data of the industry where the enterprise works during the target time period.
[0021] In an embodiment of the present application, the work data of the target employee includes: position, rank, work content, job changes, work content adjustment, performance evaluation results, and training experience. The feature vector (F_E) of the employee's personal dynamic data (E) may cover the coding of the employee's position, the quantification of changes in job responsibilities (such as the change in the complexity of the job), the numeration of performance evaluation, etc. The extraction can be based on the data in the human resources system, using classification and numerical conversion methods. Job transfers can be represented by the codes of different positions; changes in job responsibilities can be quantified by the degree of difference between the new responsibilities and the original responsibilities; performance evaluation results can be represented by scores or grades; training experience can be quantified by the number of training courses or the length of training.
[0022] In an embodiment of the present application, the historical behaviors performed by the target employee on the work computer during the target time period may be various operations performed in the internal system of the enterprise.
[0023] In the embodiment of the present application, the business planning data of the enterprise where the target employee is located includes text type and numerical type; the text type includes the company's strategic planning documents, business goal descriptions, project plan descriptions, etc. For example, descriptive statements such as "increase market share to 20% in the next three years" and "start new R&D projects to expand product lines"; numerical types may have specific business indicator values, such as expected sales, profit targets, project budgets, etc. For example, "the sales target for this year is 50 million yuan" and "the budget for a certain project is 1 million yuan." The acquisition method includes: consulting the company's strategic planning documents, annual reports, project plans, etc. These documents are usually prepared by the company's management and contain the company's long-term and short-term plans; as well as obtaining the company's strategic planning meetings, business discussion meetings, etc. Minutes; or obtaining ongoing and planned project information from the enterprise system, including project goals, timetables, and resource allocations, etc., all of which reflect the company's business planning.
[0024] In the embodiment of the present application, the industry data of the industry in which the enterprise is located includes: descriptive content market size data, growth rate, market share and other specific values in industry news reports and research reports. For example, "the development trend of emerging technologies in the industry", "the market strategy of competitors", and "the market size of a certain industry has reached 10 billion yuan", "the industry growth rate is 10%", and "the company's market share in the industry is 15%".
[0025] In the embodiments of the present application, the business planning data and industry data of the enterprise can be obtained through industry-related news websites, professional media platforms, or through public information to obtain professional market research reports, which are issued by market research institutions or consulting companies and contain detailed industry data analysis and trend forecasts, or through reports of industry conferences and seminars to collect the opinions and latest research results of industry experts; or follow industry-related social media accounts and professional forums, participate in discussions, and obtain real-time information and opinions in the industry. For example, follow the opinions of industry experts and related companies on multiple industry data analysis websites.
[0026] In another embodiment of the present application, a real-time data monitoring mechanism is established to obtain data in a timely manner, and real-time stream processing technology Apache is used to monitor the data source of market trends and competitor dynamic data in real time. For example, the real-time stream processing system receives the data source of market trend data and the data source of dynamic data of competitors in the industry in real time; for example, in Kafka Streams, new data is processed in real time by defining input streams and processing logic.
[0027] In another embodiment of the present application, when new data arrives, the search intent prediction model is updated by incremental updating.
[0028] In an embodiment of the present application, after obtaining the corresponding data, it is also necessary to: for market trends and competitor dynamic data obtained from different data sources, perform multi-source data verification by defining a verification function to determine the consistency of the data; for example, if it is numerical data, you can compare whether the difference is within a certain error range; if it is text data, you can use the semantic understanding model mentioned above to determine whether its meaning is consistent.
[0029] In an embodiment of the present application, a credibility index is established for each data source, which is determined based on factors such as the historical accuracy of the data source and the authority of the data source; when fusing multi-source data, the data is weighted according to the credibility index.
[0030] Step S120: Analyze the historical behavior to obtain the historical behavior analysis results; input the historical behavior analysis results and work data into the trained search intention prediction model to obtain the search intention prediction results of the target employees.
[0031] In the embodiment of the present application, historical behaviors are analyzed to obtain historical behavior analysis results, including: Collect historical behaviors of users at various time points within a preset time period from the system; for any historical behavior, count the number of times the historical behavior appears in the preset time period; determine whether the number of times the historical behavior appears is greater than the configured number threshold: if the number of times the historical behavior appears is greater than the configured number threshold, then treat the historical behavior as an occasional historical behavior; temporarily do not perform other processing on the occasional historical behavior; if the number of times the historical behavior appears is greater than the configured number threshold, then treat the historical behavior as a target historical behavior; from all historical behaviors within the preset time period, extract the target historical associated behavior associated with the target historical behavior; perform correlation analysis on the target historical behavior and the target historical associated behavior associated with the target historical behavior to obtain the correlation coefficient between the target historical associated behavior and the target historical behavior; input the target historical associated behavior corresponding to different target historical behaviors and any target historical behavior and the correlation coefficient into the trained behavior analysis neural network to obtain the historical behavior analysis result.
[0032] In the embodiment of the present application, historical behavior includes various behavior records of the user, such as browsed web pages, searched keywords, performed operations, etc.; each historical behavior corresponds to a time point; the target historical behavior is the frequently occurring historical behavior.
[0033] In the embodiment of the present application, historical behaviors associated with the target historical behaviors are extracted from all historical behaviors within a preset time period, including: Extract the time point of the target historical behavior; add or subtract a preset time threshold from the time point of the target historical behavior to obtain an associated time period; extract all historical behaviors in the associated time period; perform an association analysis on all historical behaviors in the associated time period and the target historical behavior to obtain the initial associated behavior of the target historical behavior; calculate the comprehensive association score of each initial associated behavior and the target historical behavior; filter the target historical associated behavior associated with the target historical behavior from the initial associated behavior based on the comprehensive association score and the configured comprehensive association score screening threshold.
[0034] In the embodiments of the present application, association analysis may be performed using techniques such as association rule mining and sequence pattern mining.
[0035] In one embodiment of the present application, the associated time period may be within half an hour before or after the time point of the target historical behavior.
[0036] In the embodiment of the present application, all historical behaviors in the associated time period are associated with the target historical behaviors, and the associated analysis results are obtained, including: Obtain target attribute data of target historical behavior and attribute data of historical behavior, as well as configured association rules; For any historical behavior, determine whether the behavior type of the historical behavior meets the behavior type association rule corresponding to the target behavior type. If so, use the historical behavior as the initial association behavior of the target historical behavior. If not, determine whether the business logic corresponding to the historical behavior conforms to the business logic association rule corresponding to the target business logic. If so, use the historical behavior as the initial association behavior of the target historical behavior. If it does not match, the number of times the historical behavior occurs within the associated time period and the time interval between the historical behavior and historical behaviors of the same behavior type with adjacent occurrence times are counted. If the number of occurrences exceeds the preset number threshold, or the time interval is less than the preset time threshold, the historical behavior is used as the initial associated behavior of the target historical behavior.
[0037] In an embodiment of the present application, any target historical behavior and any historical behavior correspond to a business logic; the business logic corresponding to any historical behavior and any target historical behavior can be specified by the user; the target attribute data includes: the target historical behavior type and the corresponding target business logic; the attribute data includes: the behavior type of the historical behavior, the frequency or frequency in the associated time period, and the corresponding business logic.
[0038] In the embodiment of the present application, the association rules include: behavior type association rules and business logic association rules.
[0039] In an embodiment of the present application, the behavior type association rules include: when the target historical behavior is a search behavior, the viewing behavior and the clicking behavior occurring within the associated time period are initial associated behaviors; when the target historical behavior is a task submission behavior, the viewing member list behavior and the task assignment behavior occurring within the associated time period are initial associated behaviors; if the historical behavior occurs after the target historical behavior, and the behavior type of the historical behavior is a feedback type, then the historical behavior is the initial associated behavior.
[0040] In the embodiment of the present application, the feedback type includes: like or favorite.
[0041] In an embodiment of the present application, all historical behaviors in an associated time period are analyzed in association with the target historical behaviors to obtain an association analysis result, that is, behaviors in which users have feedback after the target historical behaviors (i.e., like behaviors and favorite behaviors) are screened out from all historical behaviors, as well as viewing and clicking behaviors within the associated time period after search behaviors; behaviors in which member lists and task assignments are viewed after submitting tasks in a project management scenario; and behaviors that frequently occur before and after the target historical behaviors and with a short time interval.
[0042] In the embodiment of the present application, the comprehensive correlation score between each initial correlation behavior and the target historical behavior is calculated, including: For any initial correlation behavior, respectively calculate the time correlation score, behavior type correlation score, business logic correlation score, frequency correlation score, and user feedback correlation score between the initial correlation behavior and the target historical behavior; The weighted sum of the time correlation score, the behavior type correlation score, the business logic correlation score, the frequency correlation score, and the user feedback correlation score is used as the comprehensive correlation score between the initial correlation behavior and the target historical behavior.
[0043] In an embodiment of the present application, the weights of the time relevance score, behavior type relevance score, business logic relevance score, frequency relevance score and user feedback relevance score may be set to 0.2, 0.3, 0.3, 0.1, 0.1 and 0.6 respectively.
[0044] In the embodiment of the present application, the calculation formula of the comprehensive correlation score is as follows: ; in, Represents historical behavior A and target historical behavior The comprehensive correlation score of Represents historical behavior A and target historical behavior The temporal relevance score of Represents historical behavior A and target historical behavior The behavior type relevance score; Represents historical behavior A and target historical behavior Business logic relevance score; Represents historical behavior A and target historical behavior Frequency relevance score of ; Represents historical behavior A and target historical behavior The relevance score of user feedback.
[0045] In the embodiment of the present application, if ,but ;otherwise ;in, Indicates the preset time threshold, Represents the target's historical behavior time point; Indicates the time point of historical behavior A; Represents the weight of the temporal relevance score.
[0046] In the embodiment of the present application, if the historical behavior Historical behavior with target If there is a correlation in the behavior type, , otherwise 0; Indicates the weight of the behavior type relevance score.
[0047] In the embodiment of the present application, if the historical behavior Historical behavior with target If there is a connection in business logic, , otherwise 0; Indicates the weight of the business logic relevance score.
[0048] In this embodiment of the present application, if the number of times historical behavior A occurs within the associated time period is If the number of times is greater than or equal to the preset threshold, , otherwise 0; Represents the weight of the frequency relevance score.
[0049] In the embodiment of the present application, if the historical behavior Occurs after the target historical behavior, and the behavior type of the historical behavior is feedback type, then , otherwise 0; Represents the weight of the user feedback relevance score.
[0050] In an embodiment of the present application, the correlation coefficient includes: a timeliness correlation coefficient and an importance correlation coefficient; wherein the timeliness correlation coefficient is the degree of correlation between the target historical correlation behavior and the target historical behavior in terms of timeliness; the timeliness correlation coefficient is determined according to the time distance between the time point when the target historical behavior occurs and the time point when the target historical correlation behavior occurs; the smaller the time distance, the larger the timeliness correlation coefficient; different time windows can be set to evaluate timeliness, for example, the correlation behavior that occurs within one week of the target historical behavior is regarded as having a higher timeliness; the importance correlation coefficient is obtained by analyzing the frequency of the target historical correlation behavior, the correlation with the core business, and the degree of influence on the target historical behavior; the more times the target historical correlation behavior occurs, the greater the correlation with the target behavior and the greater the important influence on achieving user goals, the greater the importance correlation coefficient.
[0051] In an embodiment of the present application, by performing correlation and relevance analysis on the target historical behavior, the target historical associated behavior and the corresponding correlation coefficient associated with the target historical behavior are obtained, which are used for subsequent intention prediction and recommendation systems to provide services and suggestions that are more in line with user needs and actual conditions.
[0052] For example, the target historical behavior is "searching for product information"; by analyzing the historical behaviors before and after each "search for product information", it is found that users usually check the reviews of related products and compare the features of different products after searching; these behaviors are closely related to searching for "product information". The timeliness is judged according to the time when the behavior occurs. For example, the behavior of checking reviews within the last week is more timely. At the same time, considering that the user's job responsibility is procurement, the behavior of checking product reviews and comparing product features is of great significance to his work, so these related behaviors are judged to have a higher degree of importance. Finally, this information is applied to the recommendation system to recommend the latest reviews and comparison information of related products to users, helping users make more informed purchasing decisions.
[0053] In another embodiment of the present application, the comprehensive correlation score may be directly used as the correlation coefficient or the correlation coefficient may be calculated based on the comprehensive correlation score: .
[0054] In another embodiment of the present application, the historical behavior data may also be pre-processed, including: The similarity between different historical behaviors is calculated using the configured similarity calculation function; based on the size of the similarity threshold pre-configured for each similarity language, it is determined whether any two historical behaviors are similar historical behaviors; and similar historical behaviors are merged or not merged according to the configured filtering rules.
[0055] In an embodiment of the present application, the behavior analysis neural network adopts a deep learning model to model the user's historical behavior sequence to capture more complex time dependencies and potential patterns; the behavior analysis neural network of the present application adopts an improved RNN structure such as LSTM or gated recurrent unit (GRU), thereby avoiding the problem of gradient vanishing or gradient exploding that may occur in traditional RNN when processing long sequence data, resulting in the inability to capture long-term dependencies.
[0056] In the embodiment of the present application, the behavior analysis neural network includes the calculation of the forget gate, the input gate and the output gate; specifically includes: Forget Gate: ; Input gate, ; Output gate, ; Cell status update: ; Final cell state: ; Hidden state output: ; in, Represents the output of the forget gate at time t; determines the degree of forgetting of early browsing behaviors in the target's historical behaviors that are less relevant to the current intention; the output value is between 0 and 1, the closer to 0, the more forgotten, and the closer to 1, the less forgotten. The weight matrix representing the forget gate adjusts the degree of forgetting different behaviors according to the characteristics of the target historical behavior and the target historical associated behavior, as well as their correlation coefficients with the current task. Indicates the hidden state of the previous moment; contains the comprehensive information of the previous target historical behaviors and partial information of behaviors associated with these behaviors, affecting the forget gate's judgment of the current behavior. Represents the input at the current moment; it can be understood as the characteristic representation of the current target historical behavior. Its correlation with the previous behavior and the correlation coefficient with the current task will affect the output of the forget gate.
[0057] Represents the bias term of the forget gate; it is used to adjust the overall output tendency of the forget gate. Represents the weight associated with the association coefficient, which is used in the forget gate to adjust the effect of the association coefficient on the degree of forgetting. A vector representing the correlation coefficient between the current target historical behavior and the associated behavior; the correlation coefficient quantifies the degree of correlation between different target historical behaviors and the corresponding target historical associated behaviors and the current task. Represents the output of the input gate at time t; determines the degree of acceptance of information related to the current task in the current target's historical behavior; the output value is between 0 and 1, and the closer to 1, the more it is accepted. Represents the weight matrix of the input gate; adjusts the degree of acceptance of different behavior information according to the characteristics of the target historical behavior and the target historical associated behavior and their correlation coefficients with the current task. Represents the bias term of the input gate; used to adjust the overall output tendency of the input gate.
[0058] Represents the weight associated with the association coefficient, which is used in the input gate to adjust the impact of the association coefficient on the degree of information acceptance. Represents the output of the output gate at time t; determines how much information is output from the current cell state for subsequent intent prediction. The output value is between 0 and 1, and the closer to 1, the more output. Represents the weight matrix of the output gate; adjusts the proportion of the output information according to the characteristics of the target historical behavior and the target historical associated behavior and their correlation coefficients with the current task. Represents the bias term of the output gate; used to adjust the overall output tendency of the output gate. Represents the weight associated with the correlation coefficient, which is used in the output gate to adjust the impact of the correlation coefficient on the output information. Represents the value of the candidate cell state at time t; represents new information that may be accepted, which is affected by the target's historical behavior, the target's historical associated behavior, and the association coefficient. If the current behavior is closely related to the previous behavior and has a high degree of relevance to the current task, then the candidate cell state may contain more information that is useful for subsequent intention prediction.
[0059] A weight matrix representing the cell state; adjusts the processing of new information based on the characteristics of the target history behavior and the target history associated behavior and their correlation coefficients with the current task. Represents the bias term of the cell state, which is used to adjust the overall value of the candidate cell state. Represents the weight associated with the association coefficient, which is used to adjust the influence of the association coefficient on the candidate cell state during cell state update.
[0060] Represents the cell state at time t; it integrates the information of forgetting and input, represents the memory and update of the target historical behavior and the target historical associated behavior, and is affected by the forget gate, input gate, candidate cell state and association coefficient. Indicates the cell state at the previous moment. Contains some information about the previous target history behavior and target history associated behavior, affecting the calculation of the cell state at the current moment. Represents the weight associated with the correlation coefficient, which is used to adjust the influence of the correlation coefficient on the cell state in the final cell state; Represents the hidden state output at time t; as the final feature representation, it is used for subsequent intent prediction and recommendation, and is affected by the output gate, cell state, and association coefficient. Represents the weight associated with the association coefficient, which is used to adjust the impact of the association coefficient on the hidden state output in the hidden state output.
[0061] In actual scenarios, take the historical browsing behavior of employees as an example. Assume that when employees browse the web, the system will record these behaviors in real time. When analyzing the intentions of employees, RNN will model these historical browsing behavior sequences. The forget gate will decide which previous browsing information to forget based on the previous hidden state and current input. For example, some irrelevant early browsing behaviors may be gradually forgotten. The input gate will decide which information in the current browsing behavior to accept. For example, more attention will be paid to new web browsing related to the current task. The output gate decides how much information to output for subsequent intent prediction based on the current state. The cell state update and the calculation of the final cell state will comprehensively consider the forgotten and input information to capture long-term dependencies. The hidden state output is used as the final feature representation for subsequent intent prediction and recommendation.
[0062] In the embodiment of the present application, before the historical behavior analysis result, the work data, the business planning data and the industry data are processed, the method further includes: Preprocess data containing text information such as work data and business planning data: Use word vector models (such as Word2Vec, BERT, etc.) to convert text into vector representation and obtain a vector set. Use a pre-built semantic understanding model based on deep learning to process each vector in the vector set and obtain a semantic understanding result in order to understand the semantics of the text information. And, preprocessing is performed on the numerical data in the work data, the business planning data and the industry data: the probability density distribution of the data is estimated using kernel density estimation (KDE); for each data point, the corresponding probability density is calculated, and a probability density threshold is set. If the probability density of a data point is less than the configured probability density threshold, the data point is considered to be an outlier.
[0063] In an embodiment of the present application, before inputting the historical behavior analysis results and the work data into the trained search intent prediction model, the method further includes: The historical behavior analysis results and the work data are converted respectively to obtain a behavior feature vector and a work feature vector; and a dynamic vector of the target employee is generated based on the behavior feature vector and the work feature vector.
[0064] In an embodiment of the present application, after obtaining the above-mentioned feature vectors, it is also necessary to: use PCA to reduce the dimension of each feature vector in the vector set to obtain the feature vector after preliminary dimension reduction to reduce the dimension and amount of calculation of the data; use t-SNE to perform secondary dimension reduction on the feature vector after preliminary dimension reduction to obtain the feature vector after dimension reduction for more refined nonlinear dimension reduction.
[0065] In an embodiment of the present application, in addition to the PCA algorithm, a t-SNE (t-distributed stochastic neighbor embedding) algorithm is introduced to perform low-dimensional feature vector mapping. t-SNE represents the probability of selecting a data point as its neighbor in a high-dimensional space by calculating the conditional probability, as well as the corresponding conditional probability in the low-dimensional space; the optimization goal is minimization, and the low-dimensional vector is updated by optimization algorithms such as stochastic gradient descent.
[0066] In an embodiment of the present application, the search intent prediction model uses data enhancement technology during training. For example, for text data, random insertion, deletion, replacement and other operations can be performed to increase data diversity. At the same time, a variety of regularization methods are used, such as L1 and L2 regularization, Dropout, etc.; automated hyperparameter adjustment methods are used, such as random search. Taking random search as an example, a hyperparameter search space is defined.
[0067] In an embodiment of the present application, the method also includes: in federated learning, compressing and quantizing the model parameters of the search intent prediction model; using a quantization algorithm to map the parameter values to a finite discrete set; for example, using uniform quantization, dividing the range of parameter values into multiple intervals, each interval corresponding to a quantized value.
[0068] In an embodiment of the present application, the search intent prediction models of different departments may be different. By adopting an asynchronous federated learning algorithm, each department may upload and update the model parameters of the search intent prediction model at a different time point without waiting for all departments to synchronize. For example, weighted averaging may be used for aggregation.
[0069] In the embodiment of the present application, the historical behavior analysis results and work data are input into the trained search intention prediction model to obtain the search intention prediction results of the target employees, including: The dynamic vector of the target employee is input into the trained search intention prediction model to obtain the search intention prediction result of the target employee.
[0070] Step S130: Generate search recommendation content for target employees based on the search intention prediction results, business planning data, and industry data.
[0071] In the embodiment of the present application, based on the search intention prediction results, business planning data and industry data, the search recommendation content for the target employee is generated, including: The business planning data and the industry data are converted respectively to obtain an enterprise feature vector and an industry feature vector; the similarity between the dynamic vector and the enterprise feature vector and the industry feature vector is calculated respectively; based on the similarity calculation result, initial search recommendation content is generated; based on the search intention prediction result, the initial search recommendation content is adjusted to obtain the search recommendation content for the target employee.
[0072] In an embodiment of the present application, by calculating the similarity between search intent and enterprise vectors and industry vectors, the real-time and accuracy of search recommendations are improved, so that the recommended content more accurately reflects the current search needs of employees; and by adjusting the preliminary recommendation results based on the search intent, the comprehensiveness of the recommendation is improved.
[0073] In the embodiment of the present application, the cosine similarity may be used to calculate the similarity between the dynamic vector of the target employee and the enterprise feature vector and the industry feature vector.
[0074] In an embodiment of the present application, the business planning data and the industry data are converted separately to obtain an enterprise feature vector and an industry feature vector, both of which revolve around the search intent and are logically related to each other.
[0075] In another embodiment of the present application, generating search recommendation content for a target employee based on the search intention prediction result further includes: Keyword extraction is performed on the business planning data and the industry data respectively to obtain enterprise keywords and industry keywords; category analysis is performed on the search intention prediction results to obtain categories of the search intention prediction results; based on the categories of the search intention prediction results, historical behavior data, business planning data and the industry data are analyzed; according to the analysis results, data related to the categories of the search intention prediction results are extracted from the historical behavior data, enterprise keywords and industry keywords; the extracted data is converted into vectors; the enterprise keywords and industry keywords are converted into vectors; the similarity between the employee dynamic vector and the converted vector is calculated, and recommended content is generated based on the similarity calculation results.
[0076] For example, according to the search intent prediction results, the search intent of employees is classified. For example, if the results output by the search intent prediction model show that employees are interested in new technology-related content, then it is classified into the "new technology exploration" category; if it is a search intent related to the company's internal project process, it is classified into the "internal project focus" category, etc. Based on the classified search intent, text analysis is performed on the historical behavior data related to the intent (such as search records, browsing content, operation records, etc. under the intent category). At the same time, keyword extraction is performed in combination with the latest industry trends and relevant parts of corporate planning documents related to the search intent. For example, if the search intent is about "new technology exploration", then keywords are extracted from the historical behavior data related to new technologies, industry dynamic documents, and parts of corporate planning involving new technologies.
[0077] The extracted keyword set related to the search intent is represented as a vector. Using the word embedding technology, each keyword is mapped to a low-dimensional real number vector, and then all keyword vectors are averaged or summed to obtain the vector representation related to the search intent. For example, if the search intent is "new technology exploration", the keywords extracted from this field are converted into vectors and integrated using the word embedding technology. Similarly, the keyword set of industry dynamics and corporate planning related to the search intent is also represented as a vector. In this way, the employee vector, industry dynamics vector, and corporate planning vector based on the search intent are obtained, which all revolve around the search intent and are logically related to each other.
[0078] Calculate the similarity between the employee vector and the industry dynamics vector and enterprise planning vector related to the search intent. You can use cosine similarity. Based on the similarity calculation results, select the industry dynamics and enterprise planning content with higher similarity for recommendation. However, the selection here is based on the similarity related to the search intent. You can sort from high to low by similarity and select the top n content as the recommendation results. For example, if the employee's "new technology exploration" related vector has a high similarity with a certain industry dynamic vector about "the latest application of artificial intelligence in the industry", recommend this industry dynamic to the employee because it is consistent with his search intent (new technology exploration).
[0079] In an embodiment of the present application, initial search recommendation content is generated based on the similarity calculation results, including: according to the similarity calculation results, selecting industry dynamics and enterprise planning content with higher similarity for recommendation.
[0080] In actual applications, the selection is based on the similarity related to the search intent; the similarity can be sorted from high to low, and the first few contents are selected as the recommendation results. For example, if the employee's "new technology exploration" related vector has a high similarity with a certain industry dynamic vector about "the latest application of artificial intelligence in the industry", this industry dynamic will be recommended to the employee because it is consistent with his search intent (new technology exploration).
[0081] In another embodiment of the present application, based on the search intention prediction result, the initial search recommendation content is adjusted to obtain the search recommendation content for the target employee, including: Put recommended content that is highly relevant to the search intent at the front; put recommended content that is less relevant to the search intent at the back or filter it out directly; find more relevant recommended content from other resources of the enterprise based on the search intent; if the search intent indicates that the user is interested in a new category, introduce content in that category; consider the user's long-term preferences and historical behaviors to ensure that the recommended content not only meets the current search intent, but also meets the user's overall interests; dynamically adjust the recommendation list based on the user's immediate feedback on the recommended content (such as clicks, dwell time, etc.).
[0082] In another embodiment of the present application, after obtaining the search intent prediction result, the method further includes: Analyze the search intent prediction results; if the search intent prediction result is a functional requirement, such as looking for a certain function or solution, generate search recommendation content that meets the functional requirement; consider the role of employees in the company and recommend resources related to their responsibilities; for time-sensitive content (such as the latest news, activities, etc.), give priority to recommending the latest relevant information.
[0083] In the embodiment of the present application, the user's job information and browsing history data are obtained, and the data are deeply analyzed using neural network technology. Various features are extracted from the job information and browsing history data, such as browsed performance indicators, viewed report types, operation frequency, etc. These features are then input into a pre-trained neural network model, which can automatically learn patterns and relationships in the data, thereby inferring the user's intentions.
[0084] For example, by analyzing the team performance reports that users frequently view and the specific performance indicators they pay attention to, the model can determine that the user may be a manager and is interested in information such as team performance distribution. In the data query and display stage, when a user initiates a specific query request, the system will segment and extract keywords for the query statement, and compare it with the keywords in the embedded component library to determine the relevant data content and analysis dimensions. For example, for the query "Help me check the team performance distribution this quarter", the system will identify keywords such as "this quarter" and "team performance distribution", and determine based on the algorithm model that "result distribution" is suitable for display in a tree diagram, and "analysis by department" is suitable for display in a list, etc.
[0085] When a user performs a search operation, the system first obtains the user ID, and identifies the job information and browsing history data in the system based on the unique ID. Then, the system decomposes and infers the job and historical data to guess the user's intention, and guesses and guides key information based on this. For example, through the big data model analysis of job information and historical browsing data, if the user is judged to be a manager, and the user's next action is inferred based on the historical browsing record, the recommendation intention is formed. At the same time, the system calculates the threshold based on the data analysis results, sorts them in order from high to low (such as 0 - 1), and only displays the top three high threshold data, for example: 1. Help me check the performance distribution of the team this quarter; 2. Help me check the performance of the highest performing person in my team; 3. Help me check the performance of the lowest performing person in my team; thereby reducing the user's operating costs.
[0086] Corresponding to the above method, the embodiment of the present application also provides a device for generating employee search recommendation content, such as Figure 2 As shown, the generating device of the employee search recommendation content includes: The acquisition unit 210 is used to acquire the work data of the target employee and the historical behaviors performed on the work computer during the target time period, as well as the business planning data of the enterprise where the target employee works and the industry data of the industry where the enterprise works during the target time period; An analysis unit 220, used to analyze the historical behavior and obtain historical behavior analysis results; A prediction unit 230, configured to input the historical behavior analysis result and the work data into a trained search intention prediction model to obtain a search intention prediction result of a target employee; The generating unit 240 is used to generate search recommendation content for target employees based on the search intention prediction result, the business planning data and the industry data.
[0087] The functions of each functional unit of the device for generating employee search recommendation content provided in the above-mentioned embodiment of the present application can be realized through the above-mentioned method steps. Therefore, the specific working process and beneficial effects of each unit in the device for generating employee search recommendation content provided in the embodiment of the present application are not repeated here.
[0088] The present application also provides an electronic device, such as Figure 3 As shown, it includes a processor 310 , a communication interface 320 , a memory 330 and a communication bus 340 , wherein the processor 310 , the communication interface 320 , and the memory 330 communicate with each other via the communication bus 340 .
[0089] Memory 330, for storing computer programs; The processor 310 is used to execute the program stored in the memory 330 to implement the following steps: Obtain the target employee's work data and historical behaviors performed on his / her work computer during the target time period, as well as the target employee's company's business planning data and industry data of the company's industry during the target time period; Analyze historical behaviors and obtain historical behavior analysis results; Inputting the historical behavior analysis results and the work data into a trained search intention prediction model to obtain a search intention prediction result of a target employee; Based on the search intention prediction results, the business planning data and the industry data, search recommendation content for target employees is generated.
[0090] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.
[0091] The communication interface is used for communication between the above electronic device and other devices.
[0092] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.
[0093] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
[0094] The implementation methods and beneficial effects of the components of the electronic device in the above embodiments to solve the problems can be seen in Figure 1 The various steps in the illustrated embodiment are implemented, therefore, the specific working process and beneficial effects of the electronic device provided by the embodiment of the present application are not repeated here.
[0095] In another embodiment provided by the present application, a computer-readable storage medium is provided, in which instructions are stored. When the computer-readable storage medium is executed on a computer, the computer executes any method for generating employee search recommendation content in the above embodiments.
[0096] In another embodiment provided by the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any method for generating employee search recommendation content in the above embodiments.
[0097] Those skilled in the art will appreciate that the embodiments in the present application can be provided as methods, systems, or computer program products. Therefore, the embodiments in the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments in the present application can be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0098] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0099] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0101] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present application.
[0102] Obviously, those skilled in the art can make various changes and modifications to the embodiments in the present application without departing from the spirit and scope of the embodiments in the present application. Thus, if these modifications and variations of the embodiments in the present application fall within the scope of the claims and their equivalents in the embodiments of the present application, the embodiments of the present application are also intended to include these modifications and variations.
Claims
1. A method for generating employee search recommendation content, characterized in that: The method comprises: Obtaining the work data of the target employee and the historical behaviors performed on the work computer during the target time period, as well as the business planning data of the enterprise where the target employee works and the industry data of the industry where the enterprise works during the target time period; Analyze the historical behavior to obtain historical behavior analysis results; Inputting the historical behavior analysis results and the work data into a trained search intention prediction model to obtain a search intention prediction result of a target employee; Based on the search intention prediction results, the business planning data and the industry data, search recommendation content for target employees is generated.
2. The method according to claim 1, characterized in that Analyze the historical behavior to obtain historical behavior analysis results, including: For any historical behavior, count the number of times the historical behavior occurs in the target time period; If the number of times is greater than the configured number threshold, the historical behavior is used as the target historical behavior; Extracting target historical associated behaviors associated with the target historical behaviors from all historical behaviors of the target employee within the target time period; Performing correlation analysis on the target historical behavior and the corresponding target historical associated behavior to obtain a correlation coefficient between the target historical associated behavior and the corresponding target historical behavior; Different target historical behaviors and corresponding target historical associated behaviors and association coefficients are input into the trained behavior analysis neural network to obtain the historical behavior analysis results.
3. The method according to claim 2, characterized in that Any historical behavior corresponds to a point in time; From all historical behaviors of the target employee within the target time period, target historical associated behaviors associated with the target historical behaviors are extracted, including: Extracting the time point of the target historical behavior; Add or subtract the configured time threshold from the time point of the target historical behavior to obtain the associated time period; Performing correlation analysis on all historical behaviors of the target employee in the associated time period and the target historical behavior to obtain initial associated behaviors of the target historical behavior; Calculate the comprehensive association score between each initial association behavior and the target historical behavior; The target historical association behaviors associated with the target historical behaviors are screened out from the initial association behaviors according to the comprehensive association score and the configured comprehensive association score screening threshold.
4. The method according to claim 1, characterized in that Before inputting the historical behavior analysis results and the work data into the trained search intention prediction model, the method further includes: Convert the historical behavior analysis result and the work data respectively to obtain a behavior feature vector and a work feature vector; Based on the behavior feature vector and the work feature vector, a dynamic vector of the target employee is generated.
5. The method according to claim 4, characterized in that Inputting the historical behavior analysis results and the work data into the trained search intention prediction model includes: The dynamic vector is input into the trained search intention prediction model to obtain the search intention prediction result of the target employee.
6. The method according to claim 5, characterized in that Based on the search intention prediction result, the business planning data and the industry data, search recommendation content for target employees is generated, including: Convert the business planning data and the industry data respectively to obtain an enterprise feature vector and an industry feature vector; Calculating the similarity between the dynamic vector and the enterprise feature vector and the industry feature vector respectively; Based on the similarity calculation results, generate initial search recommendation content; Based on the search intention prediction result, the initial search recommendation content is adjusted to obtain the search recommendation content for the target employee.
7. A device for generating employee search recommendation content, characterized in that: The device comprises: An acquisition unit, used to acquire work data of a target employee and historical behaviors performed on a work computer during a target time period, as well as business planning data of the enterprise where the target employee works and industry data of the industry where the enterprise works during the target time period; An analysis unit, used to analyze the historical behavior and obtain a historical behavior analysis result; A prediction unit, used to input the historical behavior analysis result and the work data into a trained search intention prediction model to obtain a search intention prediction result of a target employee; A generating unit is used to generate search recommendation content for target employees based on the search intention prediction result, the business planning data and the industry data.
8. An electronic device, characterized in that: The electronic device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; A processor, used to implement any method described in claims 1-6 when executing a program stored in a memory.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.
Citation Information
Patent Citations
Transform-based sequence recommendation method and system
CN114780831A
Method, device and system for creating proposals and quotation of corporate tailored content based on artificial intelligence
KR102597432B1
Search Performance and User Interaction Monitoring of Search Engines
US20070265999A1
Search results based on a search history
US20160350370A1
Data storage service for personalization system
US9256761B1
Cited By
Sales target filling assisting method, device and equipment and readable storage medium
CN119203948A
Enterprise information query method and device
CN119513377A
AI digital human-based customer service platform, service method, equipment and medium
CN120508623A
