An individualized information technology consultation pushing system based on user portrait

Abstract

The application discloses a kind of personalized information technology consultation push systems based on user portrait, it is related to information technology consultation push technical field, system includes: user portrait construction module gathers user information and constructs multidimensional portrait, path prediction module predicts candidate problem evolution path according to the graph node corresponding to current consultation question, the evolution weight value of directed edge and the industry category and enterprise scale dimension in user portrait, decision reference search module searches the statistical data of peer technology decision corresponding to each node in path according to industry category and enterprise scale, and push generation module is associated with the chain push content of candidate path and peer decision statistical data and generates.This application can provide the problem push service with logical continuity and decision reference value.

Description

Technical Field

[0001] This invention relates to the field of information technology information service delivery technology, specifically a personalized information technology information service delivery system based on user profiles. Background Technology

[0002] Information technology consulting recommendation systems are widely used in scenarios such as enterprise technology selection and software development decisions. With the development of big data and artificial intelligence technologies, personalized recommendations based on user profiles have become an important direction for information technology consulting services. However, most existing consulting recommendation systems push recommendations independently based on a single user query, lacking logical connections between the recommended content, and limiting the content to the answer itself, failing to provide users with decision-making reference information.

[0003] In the prior art, the question push method, apparatus, computer-readable storage medium, and computer device with invention patent announcement number CN110457448B determine the question node corresponding to the input question in a question-and-answer database, obtain the transition probability or jump probability between the question node and other question nodes in the question-and-answer database, and determine the question to be pushed based on the transition probability or jump probability. Although this scheme achieves probability-based question push, its push is only based on the statistical transition probability between nodes. The pushed questions lack a structured association based on knowledge logic, and the pushed content is only the question itself, without any reference information to assist user decision-making. Another prior art invention patent announcement number CN119761517A proposes a system and method for question answering based on a knowledge graph-enhanced large language model. This method uses named entity recognition technology to extract the topic entities and entity types in the question, identifies the question intent based on relation classification technology and constructs a query statement, retrieves relevant fact triples through a knowledge graph, and then uses a large language model to generate the answer. Although the solution incorporates knowledge graph technology, its application scenario is single-round question-and-answer generation, rather than question sequence push in multi-round consultation scenarios. Furthermore, its push content is only the answer to a single question, and it also lacks the mining of evolutionary relationships between consultation questions and the provision of peer decision-making reference information.

[0004] The aforementioned existing technologies improve consultation push from two directions: probability-based push and question-and-answer generation. However, they address different technical issues: one focuses on the sorting of questions to be pushed, while the other focuses on the generation of answers to individual questions. Neither of these existing technologies addresses the regularity of user questions evolving along specific paths in IT consultation scenarios, nor do they provide users with peer technical decision-making references at each node of the question evolution path. However, in IT consultation scenarios, users' technical decision-making needs often permeate the entire question evolution process. As users delve deeper along the question path, they not only need to know the content of subsequent related questions but also need to obtain practical decision-making references from peers on the same questions at each path node. Neither of the aforementioned existing technologies can solve this problem. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a personalized information technology consultation and push system based on user profiles. By constructing a technical problem evolution map to predict the user's problem evolution path, and by associating and pushing statistical data of peer technical decisions at each node of the path, the system enables users to obtain consultation information along the problem evolution path while obtaining peer decision references, thereby solving the technical problems of existing consultation and push systems that lack guidance on problem evolution paths and lack decision reference information.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a personalized information technology consultation and push system based on user profiles, the system comprising: The user profile building module is used to collect the target user's industry category, company size, technical role, technology stack preference, and historical consultation records to build a user profile that includes industry category dimension, company size dimension, technical role dimension, technology stack preference dimension, and consultation history dimension. The technical problem evolution graph storage module is used to store a directed graph with information technology consultation problem types as nodes and the evolutionary relationship between problems as directed edges. The directed edges correspond to evolution weight values ​​based on the statistics of historical user consultation paths. The peer decision-making knowledge base storage module is used to store peer technical decision statistics data corresponding to each node. The peer technical decision statistics data are aggregated and stored according to industry category and enterprise size. The path prediction module is used to predict a candidate question evolution path from the directed graph based on the current node corresponding to the target user's current consultation question in the directed graph, the evolution weight values ​​of each directed edge starting from the current node, and the industry category and enterprise size dimensions in the user profile. The candidate question evolution path is an ordered sequence of nodes including the current node and at least one subsequent node. By integrating the industry category and enterprise size dimensions in the user profile to perform weighted filtering of the evolution weight values ​​of the directed edges, the predicted candidate question evolution path is accurately matched with the target user's actual technical decision-making scenario. The decision reference retrieval module is used to retrieve peer technical decision statistics data corresponding to each node in the evolution path of candidate issues from the peer decision knowledge base storage module based on the industry category dimension and enterprise size dimension in the user profile. The push generation module is used to associate the evolution path of candidate issues with the statistical data of peer technical decisions corresponding to each node in the evolution path of candidate issues, and generate chained push content. The path selection receiving module is used to receive the target user's selection operation for any subsequent node in the candidate problem evolution path; The push effect evaluation module is used to obtain the sequence of follow-up questions from the target user after receiving the chained push content.

[0007] Furthermore, the path prediction module calculates the path from the current node according to the following formula. via directed edge Reaching adjacent nodes Path preference score : in, Represents a directed edge The evolution weight value ranges from [0,1]. User profile and nodes representing target user u The semantic matching degree between the corresponding question types ranges from [0,1]. This indicates the industry category, company size, and node of the target user u. The industry suitability of the corresponding question type ranges from [0,1]. , , They are respectively , , The weighting coefficients satisfy ,and Each value ranges from [0,1]; The path prediction module selects adjacent nodes in descending order of path preference scores to generate candidate problem evolution paths. By weighting and integrating evolutionary weight values, semantic matching degree, and industry adaptability, the path selection takes into account both historical statistical patterns and the personalized characteristics of the current user, avoiding the deviation of the push path from the actual needs of the user due to relying solely on statistical probability.

[0008] Furthermore, the directed graph in the technical problem evolution graph storage module is constructed in the following manner: Information technology consulting issues are hierarchically classified into conceptual explanation, comparative analysis, selection and decision-making, implementation and deployment, and operation and maintenance management. Each issue type is used as a node, and directed edges are established based on the logical sequence between issue types. The evolution weight value is calculated by weighting the frequency of evolution along the directed edge direction in the historical user consulting sequence with the proportion of users whose problems are solved after evolving along the directed edge direction. This ensures that the weight value of the directed edge not only reflects the popularity of the path but also the effectiveness of the path, that is, whether the user's problem is actually solved after evolving along the direction.

[0009] Furthermore, the peer decision-making knowledge base storage module includes the following peer technical decision statistics: the candidate technical solution identifiers corresponding to each node, the number of times each candidate technical solution is selected within the target industry category and target enterprise size range, the selection ratio of each candidate technical solution within the target industry category and target enterprise size range, and the decision reason text corresponding to each candidate technical solution. This ensures that the content subsequently pushed to users presents decision distribution data filtered by industry and size dimensions, rather than a global and general statistical value.

[0010] Furthermore, the chained push content includes: The text of the answer to the consultation question corresponding to the current node, the node identifier sequence of each subsequent node in the candidate question evolution path excluding the current node, and the statistical data of peer technical decisions corresponding to each subsequent node; The push generation module arranges and displays the node identifier sequence according to the node order in the candidate problem evolution path, and associates and displays the peer technical decision statistics corresponding to each node with the corresponding node identifier. This allows users to see the distribution of peers' technical solution choices at each path node while browsing the evolution path. The path itself is a reference chain for users to gradually complete technical decisions.

[0011] Furthermore, in response to the selection operation, the path prediction module re-predicts the evolution path of the candidate problem by using the selected subsequent node as the new current node; The decision reference retrieval module responds to the selection operation by retrieving peer technical decision statistics data corresponding to the selected subsequent nodes based on the industry category dimension and enterprise size dimension in the user profile. The push generation module responds to the selection operation and generates updated chained push content. This allows the system to replan subsequent paths each time the user actively selects a path branch, enabling dynamic path adjustment under user guidance, rather than ending the service after a one-time static push.

[0012] Furthermore, the user profile building module is also used to assess the technical capability level of the target user based on the distribution of question types and the time taken to resolve questions in the target user's consultation history. The path prediction module adjusts the evolution weight values ​​of each directed edge starting from the current node according to the technical capability level value. The higher the technical capability level value, the greater the adjustment range of the evolution weight values ​​of the directed edges corresponding to the implementation deployment and operation and maintenance management problem nodes. This encourages novice users to evolve from conceptual explanation-based learning to selection and decision-making-based learning, while advanced users tend to evolve from selection and decision-making-based learning to implementation, deployment, and operation and maintenance management-based learning, thus adapting the path-pushing strategy to the user's level of technical knowledge.

[0013] Furthermore, the push effect evaluation module is also used to: calculate the sequence matching degree between the subsequent consultation question sequence and the candidate question evolution path, and update the evolution weight value of the corresponding directed edge in the directed graph according to the sequence matching degree; This allows the evolution weights of directed edges to be continuously optimized as the system operates for longer and real user behavior data accumulates, thus achieving adaptive evolution of the graph.

[0014] Furthermore, the push effect evaluation module is also used to: obtain the solution selection results of the target user after viewing the peer technical decision statistics, generate a decision reference validity score based on the consistency between the solution selection results and the candidate technical solutions with the highest selection ratio in the peer technical decision statistics, and update the confidence weight of the peer technical decision statistics of the corresponding node in the peer decision knowledge base storage module based on the decision reference validity score. This allows the peer decision statistics presented to users to be sorted and filtered according to confidence weights, prioritizing the display of verified decision reference information that is highly consistent with users' actual choices.

[0015] Furthermore, the user profile building module is also used to: collect consultation scenario context information of the target user, the consultation scenario context information including consultation initiation time, consultation initiation device type and consultation initiation geographical location; The path prediction module determines the scenario type based on the consultation scenario context information, and adjusts the evolution weight values ​​of each directed edge starting from the current node according to the scenario type. Different adjustment coefficients correspond to different scenario types. This allows the same user to receive differentiated push notifications during different scenarios, such as working hours and non-working hours, and on office equipment and mobile devices, thus catering to the different consultation needs of users in different scenarios.

[0016] Compared with existing technologies, this personalized information technology consultation and recommendation system based on user profiles has the following advantages: I. This invention constructs a user profile that includes dimensions such as industry category, company size, technical role, technology stack preference, and consulting history. It also creates a technical problem evolution graph with information technology consulting problem types as nodes and directed edges representing the evolutionary relationships between problems. A path prediction module integrates evolution weight values, semantic matching degree, and industry adaptability to predict the evolution path of candidate problems. A decision reference retrieval module retrieves peer technical decision statistics corresponding to each node in the path based on the industry category and company size in the user profile. Finally, a push generation module associates the candidate problem evolution path with peer decision statistics to generate chained push content. This allows users to simultaneously obtain the distribution of peers' solution choices and decision-making reasons on the same technical problem at each path node while obtaining consulting information along the problem evolution path. This solves the problem in existing consulting push systems where the lack of logical connections between push content and the limitation of push content to only the answer itself prevent users from simultaneously obtaining peer decision reference information during the consulting process.

[0017] Second, this invention receives the user's selection operation for any subsequent node in the evolution path of the candidate question through the path selection receiving module. This allows the path prediction module to re-predict the subsequent path and update the pushed content using the selected subsequent node as the new current node, thus realizing dynamic path adjustment under user guidance. Furthermore, the push effect evaluation module calculates the matching degree between the user's subsequent consultation question sequence and the candidate path to update the evolution weight value in the directed graph. At the same time, it generates an effectiveness score based on the comparison between the user's solution selection result and peer statistical data, and updates the confidence weight of the peer decision statistical data. This makes the evolution path and decision reference output by the system after multiple uses more closely match the actual technical decision-making scenario, improving the matching degree between the pushed content and the user's real needs.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a system structure block diagram of the present invention; Figure 2 This is a schematic diagram of the directed graph structure of the evolution of the technical problem of the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] See Figure 1 The system includes a user profile building module, a technical problem evolution graph storage module, a peer decision-making knowledge base storage module, a path prediction module, a decision reference retrieval module, a push generation module, a path selection and receiving module, and a push effect evaluation module. These modules work collaboratively to generate a problem evolution path matching the user's characteristics based on the user's multi-dimensional profile and technical problem evolution graph when a user initiates an IT consultation. The system also links and pushes peer decision-making statistics for each node along the path, achieving integrated provision of consultation content and decision references.

[0023] The specific implementation methods of each module are explained below: The user profile building module collects information on the target user's industry category, company size, technical role, technology stack preferences, historical consultation records, and consultation scenario context. Industry categories are subdivided according to the National Economic Industry Classification Standard, including finance, healthcare, intelligent manufacturing, internet and software, energy, and government affairs. The system determines these categories based on the company information provided during user registration or industry tags confirmed by the user during the consultation process. Company size is divided into four ranges based on the number of employees: less than 50, 50-200, 200-1000, and more than 1000. Annual revenue can also be used as a supplementary factor. Technical roles include development engineers, architects, technical managers, operations engineers, and data analysts, selected by the user upon initial system use and modifiable later. Technology stack preferences are extracted from the user's historical consultation records through keyword extraction. The system maintains a thesaurus in the information technology field, including entries for programming languages, frameworks, middleware, and cloud platforms. The system performs word frequency statistics and TF-IDF weighting on the text content of each user's consultation to generate a technology stack preference vector. Consultation history records are stored in a time-series format. Each record includes information such as the question text, question type, initiation time, resolution status, and resolution time. Consultation context information includes the consultation initiation time, initiating device type, and initiating geographical location. The initiation time is recorded using a system server timestamp and converted to the user's local time zone. The device type is identified by the User-Agent field in the HTTP request header, distinguishing between desktop and mobile devices. The geographical location is obtained through user-authorized IP address location or mobile device location information, categorized as office or non-office areas. This module also assesses the target user's technical capability level based on the distribution of question types and resolution time in their consultation history.

[0024] Specifically, the system presets a difficulty coefficient for each type of problem. The difficulty coefficient is determined based on the average time required to solve the problem and the depth of technology involved, and the difficulty coefficient ranges from 0 to 1.

[0025] The technical ability level score is initialized to 50 points. Each time a user successfully solves a problem, the level score increases according to the formula: Update, where d is the difficulty level of the problem. If the time taken to resolve the issue exceeds 1.5 times the average time taken by users of the same size in the same industry, an additional adjustment value of 2 points will be deducted.

[0026] The technical capability level value is ultimately constrained to be between 0 and 100.

[0027] This hierarchical value is used to adjust the evolution weights in subsequent path prediction.

[0028] The technical problem evolution graph storage module stores a directed graph, with information technology consulting problem types as nodes and the evolutionary relationships between problems as directed edges. Problem types are divided into five hierarchical categories based on the inherent logic of the consulting content: concept explanation, comparative analysis, selection decision, implementation and deployment, and operation and maintenance management. Each major category is further subdivided; for example, the selection decision category includes specific problem type nodes such as database selection, application framework selection, message middleware selection, and container platform selection. Nodes are uniquely identified and accompanied by labels. Directed edges represent the evolution direction from one problem type node to another. The evolution direction is determined by the logical dependencies between consulting problems; for example, users typically first understand the concept, then compare and contrast, then make a selection, then implement and deploy, and finally enter operation and maintenance management. Therefore, the graph contains directed edges from concept explanation nodes to comparative analysis nodes, from comparative analysis nodes to selection decision nodes, from selection decision nodes to implementation and deployment nodes, and from implementation and deployment nodes to operation and maintenance management nodes. Additionally, there may be lateral evolution edges within the same level; for example, within the comparative analysis category, there might be edges from microservices comparing monoliths to service mesh comparing API gateways. Each directed edge corresponds to an evolution weight value based on historical user consultation path statistics.

[0029] See Figure 2 In the diagram, circular nodes represent various problem types, directed arrows represent evolutionary relationships, and the values ​​labeled next to the arrows are the evolution weights. The evolution weights are calculated based on a weighted average of two statistical indicators: the frequency of evolution along the directed edge direction, and the percentage of user problems solved after evolution along that direction. The system considers each directed edge... Statistical analysis of historical consultation sessions from nodes Transfer to node The number of sessions is denoted as frequency. A consultation session is defined as a sequence of all consultation questions asked by the same user within a consecutive 30-minute period. If any of the following questions appear in the sequence... Adjacent If the frequency of all outgoing edges is normalized by performing a maximum-minimum normalization process, then one evolution is counted. At the same time, the final problem-solving status of each session is recorded. If a session contains... After the transition, the problem is considered resolved if the last question in the session is marked as resolved by the user or if the system detects that no new consultation on the same topic has been initiated within 30 minutes of the session ending. The calculation follows... The resolution ratio is the ratio of the number of sessions in which the problem is eventually solved to the total number of times the directed edge appears. Evolution weight value calculation formula: ,in We set it to 0.6. In this way, the evolution weight value not only reflects the popularity of the path, but also the effectiveness of the path, ensuring that popular evolution directions that can truly guide users to solve problems receive higher weights.

[0030] The peer decision-making knowledge base storage module stores statistical data on peer technical decisions corresponding to each problem type node. This data is aggregated and stored according to industry category and company size range. For each problem node, the stored data structure includes: node identifier, industry category, company size range, and a list of candidate technical solutions. Each candidate technical solution in the list includes a solution name, number of selections, selection ratio, and a textual explanation of the decision reason. The textual explanation of the decision reason comes from the decision basis explicitly entered by the user during the consultation process, as well as a summary of the reasons submitted by the user when evaluating the solution after the system pushes the solution. For example, for the node message middleware selection, under the conditions of industry category finance and company size range of 200 to 1000 people, the stored solutions might be Kafka, RabbitMQ, and Pulsar, corresponding to selection counts of 85, 40, and 25 times, and selection ratios of 56.7%, 26.7%, and 16.7%, respectively, and are associated with typical textual explanations of the decision reason, such as the high-throughput scenario requirement associated with Kafka, and the transactional message requirement associated with RabbitMQ. This module also stores the confidence weights of various decision statistics. The initial confidence weight is 1.0, and it is dynamically adjusted based on feedback from the push effect evaluation module. This is used to sort and filter statistical data during push notifications, prioritizing the display of verified reference information.

[0031] The path prediction module predicts a candidate question evolution path based on the target user's current query question's corresponding current node in the directed graph, the evolution weights of each directed edge originating from the current node, and the industry category and company size dimensions from the user profile. The prediction process is performed by calculating a path preference score. For a given current node... , and a directed edge originating from it. Reaching adjacent nodes Path preference score The following formula is used to derive: in, For directed edges The evolution weight value ranges from [0,1]. User profile and nodes representing target user u The semantic matching degree between the corresponding question types, with a value range of [0,1]; This indicates the industry category, company size, and node of the target user u. The industry suitability of the corresponding question type ranges from [0,1]. , , They are respectively , , The weighting coefficients satisfy In specific implementation , , This combined weighting allows path selection to take into account historical evolutionary statistical patterns while fully incorporating the personalized characteristics of the current user.

[0032] Semantic matching degree The calculation method is as follows: Each question type node A keyword set is pre-associated, extracted by the system from a large number of historical consultation documents based on relevant question texts for that node, containing 5 to 15 representative technical terms. User u's technology stack preference vector is generated through the aforementioned user profile construction module, denoted as... Similarly, nodes The keyword set is also transformed into a bag-of-words vector of equal length. Each dimension of the vector corresponds to a word in the vocabulary of the information technology field, and its value is a TF-IDF weight. Calculation and The cosine similarity is calculated, and if the similarity is negative, it is set to zero. The final result is... If the dimension of the technology stack preference vector in the user profile is zero, then Use the default value of 0.5.

[0033] Industry adaptability This reflects the level of attention given to this type of issue within the given industry category and company size. The system statistically analyzes the number of consultations corresponding to each industry, company size, and node in the peer decision-making knowledge base, denoted as industry as `ind`, company size range as `[ind]`, and so on. ,node The total number of consultations was The total number of consultations across all nodes within this industry scale portfolio is Industry fit calculation formula: Where M is the total number of nodes. As a smoothing factor, The higher this value, the more likely users of the same industry and size are to consult this type of question, and the higher the match with the target user's scenario.

[0034] The path prediction module calculates the current node All outgoing edges and their corresponding adjacent nodes The path preference score is used to select the top K nodes in descending order of score, and these nodes are used as the next hop in a recursive manner to form an ordered sequence of nodes containing the current node and multiple subsequent nodes, which serves as the candidate problem evolution path. The recursion depth is generally set to 3 steps to control the path length and avoid information overload. If a node has no outgoing edges, the path terminates there. The system can also provide multiple candidate paths for the user to choose from, but it mainly pushes the one with the highest score. In addition, during the path prediction process, the technical capability level value L is used to adjust the evolution weight value of directed edges. Specifically, if a node... If the issue pertains to implementation / deployment or operations / maintenance, then... Multiply by adjustment factor Adjustment factor formula: When a user's technical skill level value is higher than 50, The effective weight of this type of edge is amplified, making the path more likely to lead to the implementation and maintenance phase; when the level value is below 50, The path selection strategy focuses on conceptual explanations and selection decisions. This ensures that the path recommendation strategy is adapted to the user's level of technical knowledge.

[0035] The context of the consultation scenario also affects path prediction. The path prediction module determines the time scenario based on the consultation initiation time: if the initiation time is between 9:00 and 18:00 on a weekday, it is determined to be a working period; otherwise, it is a non-working period. If it is a working period, the weight of the directed edge evolution of the selection decision-making node is multiplied by the time scenario coefficient of 1.2; if it is a non-working period, the weight of the directed edge evolution of the concept explanation node is multiplied by the time scenario coefficient of 1.2. Regarding device type, if the consultation initiation device is a mobile device, the weight of the comparison analysis node is multiplied by a coefficient of 0.8, because users in fragmented scenarios prefer lightweight information acquisition and avoid in-depth comparison. Regarding geographical location, if the initiation location is identified as an office area, the weight of the implementation deployment node is multiplied by 1.1; if it is not an office area, the weight of the operation and maintenance management node is multiplied by 0.9. These scenario coefficients are independent of each other and can accumulate, so that the same user can receive more adaptive path push in different scenarios.

[0036] The decision reference retrieval module is responsible for retrieving statistical data on peer technical decisions corresponding to each node in the evolution path of candidate issues from the peer decision knowledge base storage module, based on the industry category and company size dimensions in the user profile. During the retrieval, it prioritizes precise matching of industry category and company size ranges. If the target user's company size is at the edge of a certain range and the data volume is insufficient, it aggregates upwards to adjacent ranges containing more samples. For example, if a user's company size is 180 people, belonging to the 50-200 person range, and if the number of nodes in this range is less than the threshold of 10, it attempts to use data from the 200-1000 person range as supplementary data, while lowering the confidence level. The retrieval results include the candidate technical solution identifier, selection frequency, selection ratio, and decision reason text for each node under the corresponding industry size, and are sorted by confidence weight, selecting the top 5 solutions with the highest confidence.

[0037] The push generation module associates the evolution path of candidate questions with the peer technical decision statistics corresponding to each node, generating chained push content. The specific chained push content includes three parts: the answer text to the current question at the current node, the node identifier sequence of each subsequent node in the candidate question evolution path (excluding the current node), and the peer technical decision statistics corresponding to each subsequent node. The answer text to the current question is obtained by searching the system's built-in consultation knowledge base, which pre-contains authoritative answers to a large number of common IT questions, or is generated in real-time by connecting to a large language model to ensure accuracy. The node identifier sequence is formed by arranging node labels in the order of the path, such as step one: microservice concept to step two: microservice architecture comparison to step three: service mesh selection to step four: Istio deployment parameter configuration. The peer technical decision statistics corresponding to each subsequent node are associated with the corresponding node identifier in the form of cards. Each card displays the solution name, a selection ratio bar chart, and a brief summary of the decision reasons. Users can click on the card to view the detailed decision reason text.

[0038] The path selection receiving module receives the target user's selection of any subsequent node in the candidate problem's evolution path. When a user clicks on a subsequent node on the interface, the system captures the event and triggers a new round of path prediction, decision reference retrieval, and push generation. The path prediction module uses the selected subsequent node as the new current node and re-executes the path preference score calculation and path generation process; the decision reference retrieval module retrieves peer decision statistics corresponding to the new node; and the push generation module generates updated chained push content, replacing the original push. This mechanism allows the system to replan subsequent paths after each user actively selects a path branch, achieving dynamic adjustments under user guidance, rather than a one-time static push.

[0039] The push notification effectiveness evaluation module acquires the sequence of subsequent inquiries from the target user after receiving the chained push notification content and performs quantitative evaluation. The evaluation is conducted at two levels: the path level and the decision reference level. At the path level, the sequence matching degree between the subsequent inquiry question sequence and the evolution path of the candidate questions is calculated. The system records the M consecutive question types initiated by the user as follows: The portion of the candidate path node sequence pushed from the current node onwards is denoted as... The longest common subsequence length is used as the criterion. The length ratio is used as the matching degree indicator, with a matching degree value range of [0,1]. If the matching degree exceeds the threshold of 0.6, it is considered that the user has actually followed the push path. The evolution weight values ​​of the corresponding directed edges in the directed graph are updated according to the matching degree. The specific update method is as follows: Each pair of directed edges If the user is If a corresponding transition exists, then the evolution frequency count of that edge is increased. This leads to the recalculation of the normalization frequency and resolution ratio, and updates. At the same time, regardless of whether the user follows the rules, the system records the actual transfer situation corresponding to this push, as new historical data accumulation, to achieve adaptive evolution of the directed graph.

[0040] At the decision-making reference level, the push notification effectiveness evaluation module obtains the solution selection results of target users after viewing peer technical decision-making statistics. When a user explicitly selects a certain technical solution in subsequent consultations, the system compares that solution with the candidate technical solution with the highest selection rate in the peer decision-making statistics. If they match, the decision-making reference effectiveness score is recorded as follows. Otherwise, record as The confidence weights of peer technical decision statistics in the peer decision knowledge base storage module are updated based on this score, according to the corresponding node, industry, and scale conditions. The confidence weights are updated using a moving average method. After a period of operation, statistical data that is highly consistent with users' actual choices will receive higher confidence weights and be prioritized for display during push notifications; statistical data that frequently deviates from users' actual choices will have lower confidence weights and their influence will gradually diminish. This makes peer decision-making reference information increasingly closer to real-world decision-making scenarios.

[0041] During operation, the user profile building module continuously accumulates user behavior data and incrementally updates factors such as technology stack preferences and technical capability levels, making the profiles increasingly accurate. The technology problem evolution graph also continuously optimizes edge weights as usage frequency increases and feedback on effects accumulates. The initial directed graph of the system can be built based on expert experience, with all directed edges given a small initial weight of 0.1, and then gradually adjusted through trial operation and the accumulation of user data.

[0042] In practical deployment, the system adopts a microservice architecture, with each module deployed independently and communicating via RESTful APIs. The technical problem evolution graph and peer decision-making knowledge base are stored using a hybrid approach of graph and relational databases. Graph databases, such as Neo4j, store nodes and directed edges, while relational databases store statistical data. Front-end push notifications can utilize WebSocket technology to ensure real-time performance.

[0043] The following example of a complete consultation process illustrates the detailed workflow of the system: User A, working in the finance industry at a company with 500 employees, serves as an architect and prefers Java, Spring Cloud, and Kubernetes. One weekday afternoon at 3 PM, the user initiated a consultation from their office using a desktop client: What are the advantages of microservice architecture? The user profiling module collected data from the request, including industry (finance), company size (200-1000 employees), technical role (architect), device type (desktop), time (workday), and location (office area), and combined this with historical consultation data to assess the user's technical skill level. .

[0044] The current consultation question text, identified by natural language processing, is classified as a microservice concept question type, belonging to a specific node under the concept explanation category. The node is coded as follows: . This node As the current node. In the technical problem evolution graph, from There are three directed edges starting from: (pointing to the comparison between microservices and monolithic architectures - nodes) ), pointing to the microservice decomposition principle (nodes) ), pointing to the microservice communication mechanism (node) Evolution weights of each edge. The values ​​are 0.72, 0.48, and 0.33, respectively.

[0045] The path prediction module begins calculating the path preference score for each edge. For example, Semantic matching degree calculate: The keyword set includes microservices, monolithic architecture, comparison, performance, complexity, and user technology stack preference vectors. The cosine similarity of the keyword vectors is 0.85. Industry adaptability In the financial industry with a scale of 200 to 1000 people, The total number of consultations for each node was 320, the total number of consultations for the industry portfolio was 1200, the total number of nodes M was set to 30, and a smoothing factor was used. , Substitute , , ,have to: User technical capability level value , This is a comparative analysis, not an implementation, deployment, or operation and maintenance management category. Not effective. In terms of scenarios, working hours provide a benefit to the selection decision-making process. For non-selective applications, the time scenario coefficient is 1; adjustments are not triggered on the device desktop; the location (office area) is effective for implementation and deployment purposes. Not implemented, coefficient 1. Therefore, the final score is 0.6682.

[0046] Recalculate , Although the microservice decomposition principle under the selection decision category belongs to the selection category, it is more design-oriented. It is actually classified as the implementation planning category and set as the implementation and deployment sub-node. , The calculated value is 0.62. Base score .because For deployment purposes, the adjustment factor is: , After adjustment Recalculated Scenario coefficient: Deployment during working hours remains unchanged; deployment on desktop devices is unaffected; deployment in office areas has a coefficient of 1.1. Final score. . For operation and maintenance management, , After adjustment , , ,Base The coefficient for location-based office area maintenance management is 0.9, but the coefficient for maintenance management in non-office areas is also 0.9. Since the office area coefficient remains unchanged, it is kept at 0.4073. After comparison... The highest score was 0.6682, so it was selected as the first jump. Then... Continuing to expand from the current node, similar calculations are performed to generate candidate paths: from microservice concept to comparison of microservices and monolithic architecture to service mesh selection to Istio deployment parameter configuration. The path length is 4 nodes.

[0047] The decision-making reference retrieval module, based on user A's financial industry background and a workforce of 200 to 1000 people, optimizes the retrieval path. , This refers to peer decision data comparing microservices and monolithic architectures, as well as service mesh selection and Istio deployment parameter configuration. For the service mesh selection node, the retrieved data shows: Istio selected 120 times (60%), Linkerd selected 55 times (27.5%), and ConsulConnect selected 25 times (12.5%), each with accompanying reasoning.

[0048] The push generation module organizes chained push notifications. The push content includes standard answers to microservice concepts, evolutionary path node sequences, and a distribution of peer choices for service mesh solutions at the service mesh selection node. After browsing the push notification, User A becomes interested in the service mesh selection and clicks on that node. The path selection receiving module captures the click event, sets the service mesh selection as the new current node, and the path prediction module re-predicts the subsequent path, deriving the path from service mesh selection to Istio control plane configuration to observability solution selection. The decision reference retrieval module retrieves the new node data and pushes the update.

[0049] In subsequent consultations with User A, the system recorded the sequence of questions they asked sequentially: service mesh selection, Istio deployment, and comparison of observability tools. The push effect evaluation module calculated the matching degree between this sequence and the candidate push path, finding a match in the first two steps and a partial match in the third step, with a matching degree of 0.75, exceeding the threshold of 0.6. Therefore, the evolution frequency and resolution rate statistics were increased for the comparison of edge microservices to service mesh selection and service mesh selection to Istio deployment, and the evolution weights were updated. Simultaneously, the user ultimately chose Istio, consistent with the highest proportion of solutions from peers, resulting in an effectiveness score of 1. The confidence weight of the Istio solution in the financial industry (200-1000 people) was updated for the service mesh selection node.

[0050] Through the specific implementation methods described above, this system deeply integrates user profiles, problem evolution maps, and peer decision-making data, achieving a personalized push service that not only guides the consultation path with logical connections but also provides practical decision-making references. The following provides implementation data for three different scenarios to demonstrate the system's operational effectiveness.

[0051] In Scenario 1, targeting medium-sized enterprises in the financial industry with 200 to 1000 employees, the system served 86 technical personnel from 42 companies during a three-month trial run, pushing chained content 1574 times. Statistics show that the average sequence matching degree between subsequent user consultation sequences and push paths was 0.81, and the proportion of users adopting peer recommendations at path nodes increased by 33% compared to random push methods. The weights of the financial industry-related edges in the technical problem evolution graph were updated a total of 328 times, showing a clear differentiation in weight distribution, with the weight of popular and effective paths increasing from around 0.3 to over 0.7.

[0052] In Scenario 2, targeting early-stage internet companies in the internet and software industry, with fewer than 50 employees and primarily development engineers, a total of 23 companies and 51 users were connected, with 947 push notifications. The average path matching score was 0.76, and the solution adoption rate increased by 28%. Since the technical capability level of early-stage teams is generally between 45 and 60, the system push paths focus more on the combination of selection decisions and implementation deployment. In consultations initiated during non-working hours on mobile devices, the proportion of push notifications increased by approximately 18% after adjusting the trigger scenario coefficient for concept explanation-type path nodes, and user feedback indicated a high degree of path relevance.

[0053] In Scenario 3, among large internet companies (over 1000 employees) and their operations engineers, there were 12 companies and 37 users, with 682 push notifications. The average path matching score reached 0.84, and the solution adoption rate increased by 40%. Due to the average user technical capability level of 79, the weight of implementation and deployment / operations management path nodes was significantly adjusted, with operations-related nodes accounting for 65% of the pushed paths. The accuracy of implementation-related path pushes was further improved in consultation scenarios initiated from office areas. After two months of feedback updates, the adoption rate of high-confidence solutions remained stable above 75%.

[0054] In summary, the system provided in this application can operate stably across multiple industries and scales. By integrating multi-dimensional user profiles, dynamic evolution maps, and peer decision-making data, it effectively provides IT consulting users with a logically continuous, personally adaptable, and decision-making reference service. The system's modules work together clearly, and the algorithm parameters have been optimized through actual testing, allowing implementation under existing engineering conditions and demonstrating good practicality and scalability.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A personalized information technology consultation and recommendation system based on user profiles, characterized in that, The system comprises: The user profile building module is used to collect the target user's industry category, company size, technical role, technology stack preference, and historical consultation records to build a user profile that includes industry category dimension, company size dimension, technical role dimension, technology stack preference dimension, and consultation history dimension. The technical problem evolution graph storage module is used to store a directed graph with information technology consultation problem types as nodes and the evolutionary relationship between problems as directed edges. The directed edges correspond to evolution weight values ​​based on the statistics of historical user consultation paths. The peer decision-making knowledge base storage module is used to store peer technical decision statistics data corresponding to each node. The peer technical decision statistics data are aggregated and stored according to industry category and enterprise size. The path prediction module is used to predict a candidate question evolution path from the directed graph based on the current node corresponding to the current consultation question of the target user in the directed graph, the evolution weight values ​​of each directed edge starting from the current node, and the industry category dimension and enterprise size dimension in the user profile. The candidate question evolution path is an ordered node sequence containing the current node and at least one subsequent node. The decision reference retrieval module is used to retrieve peer technical decision statistics data corresponding to each node in the evolution path of candidate issues from the peer decision knowledge base storage module based on the industry category dimension and enterprise size dimension in the user profile. The push generation module is used to associate the evolution path of candidate issues with the statistical data of peer technical decisions corresponding to each node in the evolution path of candidate issues, and generate chained push content. The path selection receiving module is used to receive the target user's selection operation for any subsequent node in the candidate problem evolution path; The push effect evaluation module is used to obtain the sequence of follow-up questions from the target user after receiving the chained push content.

2. The personalized information technology consultation and push system based on user profiles according to claim 1, characterized in that, The path prediction module calculates the path from the current node according to the following formula. via directed edge Reaching adjacent nodes Path preference score : in, Represents a directed edge The evolution weight value ranges from [0,1]. User profile and nodes representing target user u The semantic matching degree between the corresponding question types ranges from [0,1]. This indicates the industry category, company size, and node of the target user u. The industry suitability of the corresponding question type ranges from [0,1]. , , They are respectively , , The weighting coefficients satisfy ,and Each value ranges from [0,1]; The path prediction module selects adjacent nodes in descending order of path preference scores to generate candidate problem evolution paths.

3. The personalized information technology consultation and push system based on user profiles according to claim 1, characterized in that, The directed graph in the technical problem evolution graph storage module is constructed in the following way: Information technology consulting issues are hierarchically classified into conceptual explanation, comparative analysis, selection and decision-making, implementation and deployment, and operation and maintenance management. Each issue type is used as a node, and directed edges are established based on the logical sequence between issue types. The evolution weight is calculated by weighting the frequency of evolution along the directed edge in the historical user consulting sequence with the proportion of users whose issues are resolved after evolution along the directed edge.

4. The personalized information technology consultation and push system based on user profiles according to claim 1, characterized in that, The peer decision-making knowledge base storage module contains the following peer technical decision statistics: the candidate technical solution identifier for each node, the number of times each candidate technical solution is selected within the target industry category and target enterprise size range, the selection ratio of each candidate technical solution within the target industry category and target enterprise size range, and the decision reason text corresponding to each candidate technical solution.

5. A personalized information technology consultation and push system based on user profiles according to claim 1, characterized in that, The chained push content includes: The text of the answer to the consultation question corresponding to the current node, the node identifier sequence of each subsequent node in the candidate question evolution path excluding the current node, and the statistical data of peer technical decisions corresponding to each subsequent node; The push generation module displays the node identifier sequence in the order of nodes in the candidate problem evolution path, and associates the peer technical decision statistics data corresponding to each node with the corresponding node identifier.

6. A personalized information technology consultation and push system based on user profiles according to claim 1, characterized in that, The path prediction module responds to the selection operation by using the selected subsequent node as the new current node to re-predict the evolution path of the candidate problem. The decision reference retrieval module responds to the selection operation by retrieving peer technical decision statistics data corresponding to the selected subsequent nodes based on the industry category dimension and enterprise size dimension in the user profile. The push generation module responds to the selection operation and generates updated chained push content.

7. A personalized information technology consultation and push system based on user profiles according to claim 1, characterized in that, The user profile building module is also used to assess the technical ability level of the target user based on the distribution of question types and the time taken to resolve questions in the target user's consultation history. The path prediction module adjusts the evolution weight values ​​of each directed edge starting from the current node according to the technical capability level value. The higher the technical capability level value, the greater the adjustment range of the evolution weight values ​​of the directed edges corresponding to the implementation deployment and operation and maintenance management problem nodes.

8. A personalized information technology consultation and push system based on user profiles according to claim 1, characterized in that, The push effect evaluation module is also used to: calculate the sequence matching degree between the subsequent consultation question sequence and the evolution path of the candidate question, and update the evolution weight value of the corresponding directed edge in the directed graph according to the sequence matching degree.

9. A personalized information technology consultation and push system based on user profiles according to claim 8, characterized in that, The push effect evaluation module is also used to: obtain the solution selection results of the target user after viewing the peer technical decision statistics, generate a decision reference validity score based on the consistency between the solution selection results and the candidate technical solutions with the highest selection ratio in the peer technical decision statistics, and update the confidence weight of the peer technical decision statistics of the corresponding node in the peer decision knowledge base storage module based on the decision reference validity score.

10. A personalized information technology consultation and push system based on user profiles according to claim 1, characterized in that, The user profile building module is also used to: collect consultation scenario context information of the target user, the consultation scenario context information including consultation initiation time, consultation initiation device type and consultation initiation geographical location; The path prediction module determines the scenario type based on the context information of the consultation scenario, and adjusts the evolution weight values ​​of each directed edge starting from the current node according to the scenario type. The adjustment coefficients are different for different scenario types.

Images