A project planning and business intelligence push system based on AI and knowledge graph

Abstract

The present invention discloses a project planning and business intelligence intelligent push system based on AI and knowledge graph, which relates to the technical field of petrochemical project planning, and includes a data collection module for real-time scraping of industry data: a business intelligence analysis module, which includes a language processing unit, a knowledge graph construction unit, and a decision support unit; the language processing unit is used to extract key information according to the information of the data collection module, and transmit the key information to the knowledge graph construction unit and the decision support unit after conversion; by deepening the application of natural language processing technology, the accuracy and depth of text analysis and semantic understanding are improved, and an innovative association recommendation algorithm is used to improve the accuracy and personalization of recommendations. By analyzing the historical behavior and preferences of users, personalized recommendations can be provided even when user data is limited; and the content that users may be interested in is predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of petrochemical project planning, and more specifically, to a project planning and business intelligence intelligent push system based on AI and knowledge graphs. Background Art

[0002] The existing technology mainly focuses on the real-time scraping and efficient processing of massive data from various websites, social media, news sources, etc. This technology has been widely applied in multiple fields such as finance, e-commerce, and government affairs. After continuous optimization and iteration, it has high stability and reliability; it can support the rapid collection, cleaning, integration, and analysis of massive data, meet the all-round and multi-angle monitoring requirements for scientific and technological innovation business information, and through this technology, it is possible to achieve real-time tracking and in-depth analysis of competitors, industry trends, technological trends, etc.;

[0003] However, in the actual use process, although natural language processing (NLP) technology has made remarkable progress in text analysis, sentiment analysis, etc., it still faces challenges in deep semantic understanding and reasoning. Scientific and technological innovation business information often involves professional terms and complex contexts, and more accurate semantic understanding technology is required to extract key information and identify potential value;

[0004] Moreover, the association recommendation technology is the key to realizing personalized information push. However, the existing recommendation algorithms often have problems such as cold start and data sparsity, resulting in unsatisfactory recommendation effects and insufficient accuracy and personalization of recommendations. Summary of the Invention

[0005] Aiming at the existing deficiencies in the existing technology, the purpose of the present invention is to provide a project planning and business intelligence intelligent push system based on AI and knowledge graphs.

[0006] To achieve the above object, the present invention provides the following technical solutions on the one hand:

[0007] A project planning and business intelligence intelligent push system based on AI and knowledge graphs, including a data collection module, and the data collection module is used for real-time scraping of industry data:

[0008] A business intelligence analysis module, and the business intelligence analysis module includes a language processing unit, a knowledge graph construction unit, and a decision support unit;

[0009] The language processing unit is used to extract key information according to the information of the data collection module, and transmit the key information after conversion to the knowledge graph construction unit and the decision support unit;

[0010] The knowledge graph construction unit constructs a knowledge base based on the key information of the language processing unit, forms knowledge graph information, and transmits the knowledge graph information to the decision support unit;

[0011] The decision support unit is used to perform statistics and analysis based on the results of the language processing unit and the knowledge graph construction unit to assist in making business decisions;

[0012] An intelligent push module, the intelligent push module includes a personalized unit and a push unit, and the personalized unit is used to understand user needs according to the information of the business situation analysis module and the data collection module;

[0013] The push unit is used to push relevant information according to the judgment result of the personalized unit.

[0014] Preferably, the data collection module includes a collection unit, a processing unit and a monitoring unit;

[0015] The collection unit is used to collect raw data from different data sources;

[0016] The processing unit is used to process the raw data of the collection unit. The specific steps are as follows:

[0017] Distinguish different types of raw data;

[0018] Extract readable text from different types of raw data, perform text extraction, clean HTML content, extract useful text and remove tags;

[0019] Delete irrelevant filler words, special characters and format errors in the readable text, correct spelling mistakes and grammar errors, identify and process duplicate data;

[0020] Identify entities in the text, including personal names, place names and organization names, and standardize them, and standardize data fields such as dates, currencies and measurement units;

[0021] The monitoring unit is used to monitor the processes of the collection unit and the processing unit in real time.

[0022] Preferably, the specific working mode of the monitoring unit is as follows:

[0023] According to the formula , calculate and obtain the real-time monitoring value of the data , where N is the number of measurement features of the data, is the evaluation value of the th measurement feature of the data, is the weight coefficient of the th measurement feature of the data;

[0024] Set a threshold for the real-time monitoring value in advance. If the real-time monitoring value Q of the data is less than the threshold of the real-time monitoring value, mark the data as abnormal and delete the abnormal data.

[0025] Preferably, the specific working mode of the language processing unit is as follows:

[0026] According to the formula , calculate and obtain the comprehensive value W of the text, where E is the number of feature information in the text, and where is the evaluation value of the y-th feature information, is the weight coefficient of the y-th information;

[0027] Set a threshold for the comprehensive value in advance, compare the comprehensive value W of the text with the threshold of the comprehensive value. If the comprehensive value W of the text is greater than the threshold of the comprehensive value, transmit the text to the knowledge graph construction unit.

[0028] Preferably, the obtaining method of the evaluation values of all feature information in the text is as follows:

[0029] The feature information in the text respectively includes the keyword importance value, the sentiment tendency score, the semantic relevance score, the information novelty score and the semantic understanding score of the speech degree;

[0030] For the keyword importance value, record and mark the importance values of keywords related to the project plan in advance, and obtain the comprehensive value of all keyword importance values in the text;

[0031] The sentiment tendency score is obtained by evaluating the sentiment tendency in the text, where the positive is a positive value and the negative is a negative value, and the sentiment tendency score is obtained by obtaining the sum of all sentiment tendencies in the text;

[0032] The semantic relevance score is obtained by evaluating the semantic similarity between the text and the preset domain;

[0033] The information novelty score and the semantic understanding score of the speech degree are obtained by evaluating the in-depth semantic understanding and the novelty of the text.

[0034] Preferably, the specific working mode of the language processing unit further includes:

[0035] Obtain the similarity between each word in the text and the words in the context, and then take the average value of all similarities to obtain the context adjustment factor T;

[0036] According to the formula , obtain the adaptive adjustment factor, where G is the term weight value, J is the domain matching score, and A is a preset parameter used to balance the relative importance of the term weight and the model matching degree in the calculation;

[0037] Obtain the continuous adjustment factor Y by dividing the usage performance improvement score by the maximum performance improvement score;

[0038] According to the formula , obtain the adjusted comprehensive value , and use the adjusted total value to replace the comprehensive value W for evaluation.

[0039] Preferably, the specific working mode of the knowledge graph construction unit is as follows:

[0040] Obtain the text transmitted by the language processing unit;

[0041] Use named entity recognition (NER) technology to annotate the entities in the text, and apply relation extraction technology to determine the semantic connections between entities;

[0042] Define the ontology and schema, establish the structure of the knowledge graph, map the extracted entities and relationships to the predefined ontology, and use a graph database or other storage solutions to construct the knowledge graph;

[0043] Add detailed attribute and classification information to the entities, use graph reasoning technology to discover new knowledge and connections, and complete the construction of the knowledge graph.

[0044] Preferably, the specific working mode of the decision support unit is as follows:

[0045] Obtain the market demand, market growth rate, and technology adaptability of the user enterprise according to the text in the knowledge graph construction unit and the language processing unit;

[0046] Obtain the market potential evaluation value H by multiplying the market demand by the market growth rate and then by the technology adaptability;

[0047] Obtain the return on investment Z of the user enterprise by subtracting the investment cost from the net profit of the user enterprise and then dividing by the investment cost;

[0048] Obtain the risk evaluation value C by multiplying the potential loss of the enterprise by the occurrence probability and then dividing by the risk tolerance;

[0049] Standardize and calculate the sum of the market potential evaluation value H, the return on investment Z of the enterprise, and the risk evaluation value C to obtain the total decision score N;

[0050] Preset an acceptance threshold for the total decision score. If the current total decision score N is greater than the acceptance threshold, mark the strategy of the current enterprise as feasible; otherwise, adjustment is required.

[0051] Preferably, the specific working mode of the personalization unit is as follows:

[0052] Obtain the number of types of the content browsed by the user and the number of times each content is browsed, calculate the sum of each browsed content of the user multiplied by the corresponding number of browsing times to obtain the preference value V of the user;

[0053] Obtain the feature vectors of the content to be pushed for each type;

[0054] According to the formula , calculate and obtain the preference weight of the user for each type , where represents the preference weight for the th type, is the total number of content categories, is the preference value of the user for the nd type of content, where is any integer from 1 to , representing different content categories;

[0055] According to the formula , obtain the content score of each content, where is the value of each type of content on the th feature, represents the preference weight of the th feature;

[0056] Obtain the freshness B of the content;

[0057] According to the formula , obtain the final recommendation score , set a threshold for the recommendation score, and transmit the push content with a recommendation score higher than the threshold to the said push unit.

[0058] Preferably, the specific working mode of the push unit is as follows:

[0059] Enrich the recommendation list of the user with the content with a high recommendation score and recommend it to the user, and at the same time collect the feedback and participation of the user on the push content.

[0060] Working principle

[0061] By deepening the application of natural language processing technology to improve the accuracy, depth of text analysis and semantic understanding, and innovating relevant recommendation algorithms to enhance the accuracy and personalization of recommendations. By analyzing users' historical behaviors and preferences, personalized recommendations can be provided even when user data is limited; and predict the content that users may be interested in, even when the data is insufficient, which is beneficial to providing customized information and analysis according to users' specific needs and preferences, enhancing the user experience; combining real-time market analysis and user portraits to ensure that the pushed content is highly relevant to users' interests and needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a system block diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] Refer to Figure 1 : A project planning and business intelligence push system based on AI and knowledge graph, including a data collection module, and the data collection module is used to capture industry data in real time: It should be noted that in this embodiment, the industry data specifically includes market data: price index: market prices of various commodities and resources; trading volume: an indicator of market trading activity; company information: corporate financial reports: quarterly and annual financial reports of listed companies; corporate news: major events of enterprises, new product launches, partnership relationships, etc.; industry reports: industry analysis: analysis reports on the development trends of specific industries; market research: market research results and consumer behavior analysis; policy updates: laws and regulations: newly introduced or updated laws and regulations related to the industry; policy guidance: government guidance opinions and policy support for the development of the industry; It should also be noted that data can be comprehensively captured from multiple channels such as websites, social media, and news sources to achieve real-time monitoring and rapid response to industry dynamics and information;

[0064] It should be noted that the existing technology mainly focuses on the real-time capture and efficient processing of massive data from various websites, social media, news sources, etc. This technology has been widely applied in multiple fields such as finance, e-commerce, and government affairs. After continuous optimization and iteration, it has high stability and reliability; it can support the rapid collection, cleaning, integration, and analysis of massive data, meet the all-round and multi-angle monitoring requirements for scientific and technological innovation business information, and through this technology, real-time tracking and in-depth analysis of competitors, industry dynamics, technology trends, etc. can be achieved;

[0065] However, in the actual use process, although natural language processing (NLP) technology has made remarkable progress in text analysis, sentiment analysis, etc., it still faces challenges in deep semantic understanding and reasoning. Scientific and technological innovation business information often involves professional terms and complex contexts, and more accurate semantic understanding technology is needed to extract key information and identify potential value;

[0066] Moreover, the associated recommendation technology is the key to realizing personalized information push. However, existing recommendation algorithms often have problems such as cold start and data sparsity, resulting in unsatisfactory recommendation effects, insufficient accuracy and personalization of recommendations;

[0067] This technical solution can solve the above problems by deepening the application of natural language processing technology to improve the accuracy and depth of text analysis and semantic understanding, and innovating associated recommendation algorithms to improve the accuracy and personalization of recommendations;

[0068] A business intelligence analysis module, where the business intelligence analysis module includes a language processing unit, a knowledge graph construction unit, and a decision support unit;

[0069] The language processing unit is used to extract key information based on the information of the data collection module, and transmit the key information after conversion to the knowledge graph construction unit and the decision support unit; it should be noted that in this embodiment, the information of the data collection module also includes news reports, social media content, professional analysis reports, enterprise announcements, and policy documents, etc. It should also be noted that the extracted key information includes topics, emotions, keywords, and summaries, etc.;

[0070] The knowledge graph construction unit constructs a knowledge base according to the key information of the language processing unit, forms knowledge graph information, and transmits the knowledge graph information to the decision support unit; it should be noted that the knowledge graph construction unit provides background knowledge for the decision support unit to help conduct more accurate market analysis, competition analysis, and risk assessment;

[0071] The decision support unit is used to perform statistics and analysis based on the results of the language processing unit and the knowledge graph construction unit to assist in making business decisions; it should be noted that it can understand professional terms and complex contexts more accurately, thereby extracting more accurate key information, corresponding to overcoming the limitations of the existing technology, and providing more comprehensive, in-depth, and accurate business intelligence services;

[0072] An intelligent push module, where the intelligent push module includes a personalization unit and a push unit, and the personalization unit is used to understand user needs based on the information of the business intelligence analysis module and the data collection module;

[0073] The pushing unit is used to push relevant information according to the judgment result of the personalization unit. It should be noted that in this embodiment, the relevant information includes market trends, industry dynamics, economic indicator analysis, etc. It should also be noted that by analyzing the user's historical behavior and preferences, personalized recommendations can be provided even when the user data is limited; and the content that the user may be interested in can be predicted, even when the data is insufficient, which is beneficial to providing customized information and analysis according to the specific needs and preferences of the user, enhancing the user experience; combining real-time market analysis and user portraits to ensure that the pushed content is highly relevant to the user's interests and needs.

[0074] As an optional embodiment: The data acquisition module includes an acquisition unit, a processing unit, and a monitoring unit; it should be noted that in the big data environment, the quality and accuracy of data directly affect the subsequent analysis and application effects. However, due to the diversity of data sources, there are a large amount of noise, redundancy, and inconsistent information in the data, which will affect the subsequent judgment. This technical solution uses data cleaning and integration technologies to ensure the quality and consistency of data, providing a reliable data basis for subsequent text analysis and knowledge graph construction;

[0075] The acquisition unit is used to collect raw data from different data sources; it should be noted that in this embodiment, specifically, web crawler technology is used to crawl web data, third-party data sources are integrated through APIs, and data is imported from the enterprise internal system, such as CRM, ERP, etc., supporting multiple data formats, including text, JSON, XML, etc.;

[0076] The processing unit is used to process the raw data of the acquisition unit, and the specific steps are as follows:

[0077] Distinguish different types of raw data; it should be noted that in this embodiment, the data formats are text, PDF, HTML, etc.;

[0078] Extract readable text from different types of raw data, perform text extraction, and clean the HTML content, extract useful text and remove tags;

[0079] Delete irrelevant filler words, special characters, and formatting errors in the readable text, correct spelling mistakes and grammar errors, identify and process duplicate data;

[0080] Identify entities in the text, including personal names, place names, and organization names, and standardize them, and standardize the data fields of dates, currencies, and measurement units; it should be noted that in this embodiment, specifically, NLP technology is applied for entity recognition, and a standardization list or database is used to map the same entity with different expressions;

[0081] It should also be noted that the data extracted from diverse data sources can be ensured to be not only clean and accurate, but also capable of supporting subsequent analysis, knowledge graph construction, and personalized services in a structured and useful manner;

[0082] The monitoring unit is used to monitor the processes of the acquisition unit and the processing unit in real time.

[0083] As an optional embodiment: The specific working mode of the monitoring unit is as follows: It should be noted that in actual use, data deviations may occur in the collected data and during the data processing process, and the data may be affected by other factors such as timeliness, resulting in deviations in subsequent judgments and processing. This technical solution can monitor the processes of the acquisition unit and the processing unit in real time and eliminate abnormal data information;

[0084] According to the formula , the real-time monitoring value of the data is calculated and obtained , where N is the number of measurement features of the data, is the evaluation value of the th measurement feature of this data, is the weight coefficient of the th measurement feature of this data;

[0085] It should be noted that the real-time monitoring value is obtained by comprehensively calculating each measurement feature in the data and is used to evaluate whether there is an abnormality in this data, which is beneficial to subsequent elimination of abnormal information, ensuring the quality and consistency of the data in the entire system, and providing a reliable data basis for subsequent text analysis and knowledge graph construction;

[0086] It should be noted that in this embodiment, the number of measurement features is 5, which are the integrity value, accuracy value, consistency value, timeliness value, and uniqueness value in sequence. Correspondingly, i is an integer from 1 to N, can be , can be , corresponding to the integrity value, accuracy value, consistency value, timeliness value, and uniqueness value in sequence;

[0087] It should also be noted that in this embodiment, the integrity value can be obtained by comprehensively dividing the scores of each feature in the text data by the number of features. For example, it can be obtained by dividing the sum of the number of keyword occurrences and the number of sentences by 2. The integrity value can be equivalent to the text richness score;

[0088] The accuracy value can be obtained by , the accuracy value is used to represent the error detection rate of the text, where the number of error texts is the number of error records found through fact-checking or other verification means;

[0089] The consistency value represents the format consistency score of the text, obtained by dividing the number of texts that meet the text requirements by the total number of texts;

[0090] The timeliness value is obtained by dividing the difference between the current time and the text release time by the relevance time window. The relevance time window is the time period during which the text content remains highly relevant. If the text release time is within the window from the current time, the timeliness score is relatively high;

[0091] The way to obtain uniqueness can be , where text similarity detection methods can be used to identify duplicate texts;

[0092] In this embodiment, The initial values of are all 0.2, respectively representing the influence degree of each metric feature, In this embodiment, the values can be adjusted according to the feedback of the staff, specifically 0.198, 0.203, 0.106, 0.297, and 0.196;

[0093] Set a threshold for the real-time monitoring value in advance. If the real-time monitoring value Q of this data is less than the threshold of the real-time monitoring value, then mark this data as abnormal and delete the abnormal data.

[0094] As an alternative embodiment: The specific working method of the language processing unit is as follows: It should be noted that although natural language processing (NLP) technology has made remarkable progress in text analysis, sentiment analysis, etc., it still faces challenges in deep semantic understanding and reasoning. The scientific and technological innovation business information often involves professional terms and complex contexts, and more accurate semantic understanding technology is needed to extract key information and identify potential value;

[0095] According to the formula , calculate and obtain the comprehensive value W of this text, where E is the number of feature information in the text, where is the evaluation value of the y-th feature information, is the weight coefficient of the y-th information;

[0096] It should be noted that the comprehensive value W of the text is obtained by comprehensively calculating the scores of each type of feature information in the text, which can comprehensively measure the key information in the text and identify potential value, facilitating the subsequent screening of texts with key information and potential value, enabling a more accurate understanding of the semantics in the text, and making the subsequent knowledge graph construction more accurate;

[0097] It should be noted that in this embodiment, the value of y is an integer from 1 to E. The characteristic information can be the keyword importance value, the sentiment tendency score, the semantic relevance score, the information novelty score, the semantic understanding score of the language degree, and the comprehensive text value score. Correspondingly, the weight coefficients of each characteristic information can be , corresponding to the above characteristic information in sequence respectively;

[0098] It should also be noted that in this embodiment, The initial values of are all 0.2, corresponding to the influence degree of each characteristic information on the comprehensive value. In the actual use process, The values of can be adjusted according to the evaluation of the staff and historical data. In this embodiment, The values of can be 0.158, 0.243, 0.176, 0.224, and 0.199;

[0099] It should also be noted that by considering keyword importance, sentiment tendency, semantic relevance, information novelty, and in-depth semantic understanding, this method can comprehensively evaluate multiple key aspects of the text; the comprehensive score can balance different advantages and disadvantages of the text. For example, a text may have a high score in sentiment tendency but a low score in information novelty. If the traditional calculation method is used, the comprehensive score may deviate, but the comprehensive score calculated above can reflect this balance; this calculation method can adapt to different text types and fields. By adjusting the weights and measurement criteria, it can be optimized for specific types of texts (such as scientific and technological business information); by comprehensively considering multiple dimensions, this method can more accurately identify the key information and potential value of the text, especially in the case of involving professional terms and complex contexts;

[0100] A threshold value of the comprehensive value is set in advance, and the comprehensive value W of the text is compared with the threshold value of the comprehensive value. If the comprehensive value W of the text is greater than the threshold value of the comprehensive value, the text is transmitted to the knowledge graph construction unit. It should be noted that the comprehensive value W is used to comprehensively evaluate the potential value in the text. If the comprehensive value W is greater than the threshold value of the comprehensive value, it means that the text has a certain value and can be used for subsequent knowledge graph construction; the potential value is used to measure whether the data in the text is relevant to the subsequent project planning.

[0101] As an optional embodiment: The acquisition method of the evaluation values of all characteristic information in the text is as follows:

[0102] The characteristic information in the text respectively includes the keyword importance value, the sentiment tendency score, the semantic relevance score, the information novelty score, and the semantic understanding score of the language degree. It should be noted that in actual use, the characteristic information can be increased or decreased;

[0103] For the importance value of keywords, record in advance the importance values of keywords related to marking and project planning, and obtain the comprehensive value of the importance values of all keywords in the text; it should be noted that the importance values of keywords related to project planning are obtained through prior evaluation by staff;

[0104] The sentiment tendency score is obtained by evaluating the sentiment tendency in the text, where positive is a positive value and negative is a negative value, and the sentiment tendency score is obtained by summing up all the sentiment tendencies in the text;

[0105] The semantic relevance score is obtained by evaluating the semantic similarity between the text and a preset domain; it should be noted that in this embodiment, the preset domain is set through evaluation by staff;

[0106] The information novelty score and the degree of semantic understanding score are obtained by evaluating the deep semantic understanding and the novelty of the text. It should be noted that the novelty score is obtained through the Jaccard similarity between the text and the documents in the existing document set, and the degree of semantic understanding score is obtained through the depth of semantic understanding of each logical unit in the text.

[0107] As an optional embodiment: The specific working mode of the language processing unit further includes: It should be noted that, however, in the actual use process, when calculating semantic relevance and deep semantic understanding, the context information of the text needs to be considered to more accurately capture the meaning of the text, and for texts in specific fields, such as scientific and technological business information, it may be necessary to customize the understanding of professional terms and industry-specific contexts. Over time, it is necessary to continuously collect feedback and optimize the model to adapt to the evolution of the language. This technical solution can solve the above problems;

[0108] Obtain the similarity between each word in the text and the context words, and then take the average of all similarities to obtain the context adjustment factor T; it should be noted that the similarity used here can be point mutual information or cosine similarity, both of which are indicators to measure whether two words are semantically similar;

[0109] According to the formula , obtain the adaptive adjustment factor, where G is the term weight value, J is the domain matching score, and A is a preset parameter used to balance the relative importance of the term weight and the model matching degree in the calculation; it should be noted that the term weight value is obtained by determining professional terms related to a specific domain and calculating the frequency of each term appearing in the domain documents, and the domain matching score is obtained by staff using text matching techniques (such as vector space model, semantic similarity calculation) to evaluate the matching degree of the text with the domain model, specifically an integer from 1 to 5, and the preset parameter A is from 0 to 1, and in this embodiment, it is 0.495;

[0110] The continuous adjustment factor Y is obtained by dividing the performance improvement score by the maximum performance improvement score. It should be noted that the performance improvement score is obtained by determining a baseline model for comparing performance improvement, using appropriate evaluation metrics (such as accuracy, recall, F1-score) to evaluate the model performance, and calculating the difference between the performance metrics of the new model and the baseline model as the performance improvement score.

[0111] Maximum performance improvement score: By defining an ideal model or a standard of optimal performance, which can be an assumption based on domain knowledge or technological limits, evaluate the maximum value that the model performance can reach under the current technology and data conditions, and calculate the gap between the current model performance and the ideal model performance as the maximum performance improvement score;

[0112] According to the formula , the adjusted comprehensive value is obtained , comprehensive value is used to replace the comprehensive value W for evaluation, which can more accurately reflect the comprehensive value of the text, especially when dealing with texts that require in-depth semantic understanding and domain-specific knowledge.

[0113] As an optional embodiment: The specific working mode of the knowledge graph construction unit is as follows:

[0114] The text transmitted by the language processing unit is obtained;

[0115] Use named entity recognition (NER) technology to label the entities in the text, and apply relation extraction technology to determine the semantic connections between entities. It should be noted that key entities (companies, technologies, and projects) and their relationships (invested in and applied to) are identified from the text;

[0116] Define the ontology and schema, establish the structure of the knowledge graph, map the extracted entities and relationships to the predefined ontology, and use a graph database or other storage solutions to construct the knowledge graph. It should be noted that a structured knowledge base is created to represent the extracted entities and relationships in the form of a graph;

[0117] Add detailed attribute and classification information to the entities, use graph reasoning technology to discover new knowledge and connections, and complete the construction of the knowledge graph. It should be noted that the knowledge graph is enriched by supplementing attributes, adding type information, performing graph reasoning, etc., and applied to business scenarios.

[0118] As an optional embodiment: The specific working mode of the decision support unit is as follows:

[0119] Obtain the market demand, market growth rate, and technology adaptability of the user enterprise according to the text in the knowledge graph construction unit and the language processing unit;

[0120] Obtain the market potential evaluation value H by multiplying the market demand by the market growth rate and then by the technology adaptation degree;

[0121] Obtain the return on investment Z of the user enterprise by subtracting the investment cost from the net profit of the user enterprise and then dividing by the investment cost; it should be noted that the net profit and investment cost of the enterprise are obtained from the text in the knowledge graph construction unit and the language processing unit;

[0122] Obtain the risk assessment value C by multiplying the potential loss of the enterprise by the occurrence probability and then dividing by the risk tolerance; it should be noted that the potential loss, occurrence probability and risk tolerance of the enterprise are obtained from the text in the knowledge graph construction unit and the language processing unit

[0123] Standardize the market potential evaluation value H, the return on investment Z of the enterprise and the risk assessment value C and then calculate the sum to obtain the total decision score N; it should be noted that standardization means converting them into standardized scores between 0 and 1, which can be achieved by the maximum and minimum values of each index; it should be noted that the total decision score is used to measure the rationality of the current enterprise decision of the user, obtained by integrating market factors, investment returns and risks, and is conducive to subsequent decision support;

[0124] Set an acceptance threshold for the total decision score in advance. If the current total decision score N is greater than the acceptance threshold, mark the strategy of the current enterprise as feasible; if not, adjustment is required. It should be noted that the specific adjustment method can be to identify the key factors affecting H, Z, C according to the in-depth analysis provided by the knowledge graph construction unit and the language processing unit, and formulate targeted improvement measures;

[0125] Risk management: If the risk assessment value C is high, explore risk mitigation strategies such as diversification of investments, purchase of insurance or adoption of hedging strategies;

[0126] Market strategy optimization: If the market potential evaluation value H is insufficient, study market segmentation, target customer groups or product positioning to increase market demand and technology adaptation degree;

[0127] Cost-benefit analysis: If the return on investment Z is not ideal, optimize the cost structure, improve operational efficiency or adjust the pricing strategy.

[0128] As an optional embodiment: The specific working mode of the personalized unit is as follows:

[0129] Obtain the number of types of the content browsed by the user and the number of times each piece of content is browsed, calculate the sum of each piece of content browsed by the user multiplied by the corresponding number of browsing times to obtain the preference value V of the user; it should be noted that calculating the preference frequency of the user for different types of content to obtain the preference value V, the number of types of the content browsed by the user and the number of times each piece of content is browsed are obtained through the text in the language processing unit, and the preferences of the user can be evaluated based on the historical data of the user; assist in completing the analysis of the user, and by analyzing the historical behavior and preferences of the user, personalized recommendations can be provided even when the user data is limited;

[0130] Obtain the feature vectors of the content to be pushed for each type; it should be noted that these features may include the category label of the content, keyword weight, release time, etc.;

[0131] According to the formula , calculate and obtain the preference weight of the user for each type , where represents the preference weight for the th type, is the total number of content categories, is the preference value of the user for the th type of content, where is any integer from 1 to , representing different content categories;

[0132] According to the formula , obtain the content score of each piece of content , where is the value of each type of content on the th feature, represents the preference weight of the th feature;

[0133] Obtain the freshness B of the content; it should be noted that in this embodiment, the freshness B is obtained by subtracting the release time from the current time and then dividing by the time decay constant. In this embodiment, the time decay constant can be 1.023; and the freshness represents the value of the content decaying over time;

[0134] According to the formula , obtain the final recommendation score , set a threshold for the recommendation score, and transmit the push content with a recommendation score higher than the threshold to the push unit. It should be noted that combining the content score and the freshness score to obtain the final recommendation score can be more accurate. According to the calculated recommendation score, select the content with the highest score or above a certain threshold to be pushed to the user.

[0135] As an alternative embodiment: The specific working mode of the pushing unit is as follows:

[0136] Enrich the user's recommendation list with content having a high recommendation score and recommend it to the user. At the same time, collect the feedback and engagement of the user with the pushed content.

[0137] It should be noted that specifically, it can be to collect the feedback data of the user on the pushed content, including engagement indicators such as click-through rate, reading time, sharing times, and user stay time. Use data analysis tools to monitor the performance of the pushing activity, track key indicators in real time. According to the collected data, evaluate the user engagement and satisfaction of the pushed content. Use A / B testing to compare the effects of different pushing strategies to determine which method can attract users more. Analyze the direct feedback of users, such as comments, ratings, or questionnaires, to understand the specific needs and preferences of users, identify common themes and patterns in user feedback as the basis for optimization. Adjust the pushing frequency, time, content type, etc. according to the evaluation results to improve user engagement.

[0138] Working principle: By deepening the application of natural language processing technology to improve the accuracy, depth of text analysis and semantic understanding, and innovating the associated recommendation algorithm to improve the accuracy and personalization of recommendations. By analyzing the historical behavior and preferences of users, personalized recommendations can be provided even when user data is limited; and predict the content that users may be interested in, even when the data is insufficient, which is beneficial to provide customized information and analysis according to the specific needs and preferences of users, enhancing the user experience; combine real-time market analysis and user portraits to ensure that the pushed content is highly relevant to the interests and needs of users.

[0139] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be pointed out that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of this template.

Claims

1. An AI and knowledge graph-based project planning and business intelligence push system, characterized in that, It includes a data acquisition module for capturing industry data in real time and extracting readable text from the industry data; A business intelligence analysis module, including a language processing unit, a knowledge graph construction unit, and a decision support unit; The language processing unit is used to extract key information based on the information of the data acquisition module, and transfer the key information to the knowledge graph construction unit and the decision support unit after transformation; The knowledge graph construction unit constructs a knowledge base based on the key information of the language processing unit, forms knowledge graph information, and transfers the knowledge graph information to the decision support unit; The decision support unit is used to perform statistics and analysis based on the results of the language processing unit and the knowledge graph construction unit to assist in making business decisions; An intelligent push module, including a personalization unit and a push unit, where the personalization unit is used to understand user needs based on the information of the business intelligence analysis module and the data acquisition module; The push unit is used to push relevant information based on the judgment result of the personalization unit; The specific working method of the language processing unit is as follows: According to the formula , calculate the comprehensive value W of this text, where E is the number of feature information in the text, and where is the evaluation value of the y-th feature information, is the weight coefficient of the y-th information; The feature information in the text respectively includes keyword importance value, sentiment tendency score, semantic relevance score, information novelty score, and language degree semantic understanding score; Obtain the similarity between each word in the text and the context words, and then take the average value of all similarities to obtain the context adjustment factor T; According to the formula , the adaptive adjustment factor U is obtained, where G is the term weight value, J is the domain matching score, and A is a preset parameter used to balance the relative importance of the term weight and the domain matching degree in the calculation; Divide the performance improvement score by the maximum performance improvement score to obtain the continuous adjustment factor Y; According to the formula , obtain the adjusted comprehensive value , and use the adjusted comprehensive value to replace the comprehensive value W for evaluation; Preset a threshold for the comprehensive value in advance. If the comprehensive value W of this text is greater than the threshold of the comprehensive value, then transfer this text to the knowledge graph construction unit.

2. The project planning and business intelligence push system based on AI and knowledge graph according to claim 1, wherein The data acquisition module includes a collection unit, a processing unit, and a monitoring unit; The collection unit is used to collect raw data from different data sources; The processing unit is used to process the raw data of the collection unit. The specific steps are as follows: Distinguish different types of raw data; Extract readable text from different types of raw data, perform text extraction, clean HTML content, extract useful text and remove tags; Delete irrelevant filler words, special characters, and formatting errors in the readable text, correct spelling mistakes and grammar mistakes, and identify and process duplicate data; Identify entities in the text, including personal names, place names, and organization names, and standardize them, and standardize data fields of dates, currencies, and measurement units; The monitoring unit is used to monitor the processes of the collection unit and the processing unit in real time.

3. An item planning and business intelligence push system based on AI and knowledge graph according to claim 2, characterized in that, The specific working method of the monitoring unit is as follows: According to the formula , calculate to obtain the real-time monitoring value of the said data , where N is the number of measurement features of the data, is the evaluation value of the th measurement feature of this data, is the weight coefficient of the th measurement feature of this data; Preset a threshold for the real-time monitoring value in advance. If the real-time monitoring value Q of this data is less than the threshold of the real-time monitoring value, then mark this data as abnormal and delete the abnormal data.

4. An intelligent project planning and business intelligence push system based on AI and knowledge graph according to claim 1, characterized in that The acquisition method of the evaluation values of all the feature information in this text is as follows: For the keyword importance value, record the importance values of the marked and project planning related keywords in advance, and obtain the comprehensive value of all the keyword importance values in the text; The sentiment tendency score is obtained by evaluating the sentiment tendency in the text, where positive is a positive value and negative is a negative value, and the sentiment tendency score is obtained by obtaining the sum of all the sentiment tendencies in the text; The semantic relevance score is obtained by evaluating the semantic similarity between the text and a preset domain; The information novelty score and the semantic understanding score are obtained by evaluating the in-depth semantic understanding and the novelty of the text.

5. An item planning and business intelligence push system based on AI and knowledge graph according to claim 1, characterized in that, The specific working mode of the knowledge graph construction unit is as follows: Obtain the text transmitted by the language processing unit; Use named entity recognition (NER) technology to annotate entities in the text, and apply relation extraction technology to determine the semantic connections between entities; Define the ontology and schema, establish the structure of the knowledge graph, map the extracted entities and relationships to the predefined ontology, and use a graph database or other storage solutions to construct the knowledge graph; Add detailed attribute and classification information to the entities, use graph reasoning technology to discover new knowledge and connections, and complete the construction of the knowledge graph.

6. An item planning and business intelligence push system based on AI and knowledge graph according to claim 5, characterized in that, The specific working mode of the decision support unit is as follows: Obtain the market demand, market growth rate, and technology adaptability of the user enterprise according to the text in the knowledge graph construction unit and the language processing unit; Obtain the market potential evaluation value H by multiplying the market demand by the market growth rate and then by the technology adaptability; Obtain the return on investment Z of the user enterprise by subtracting the investment cost from the net profit of the user enterprise and then dividing by the investment cost; Obtain the risk evaluation value C by multiplying the potential loss of the enterprise by the occurrence probability and then dividing by the risk tolerance; Standardize and calculate the sum of the market potential evaluation value H, the return on investment Z of the enterprise, and the risk evaluation value C to obtain the total decision score N; Preset an acceptance threshold for the total decision score in advance. If the current total decision score N is greater than the acceptance threshold, mark the strategy of the current enterprise as feasible; otherwise, adjustment is required.

7. An item planning and business intelligence push system based on AI and knowledge graph according to claim 1, characterized in that, The specific working mode of the personalization unit is as follows: Obtain the number of types of browsing content of the user and the number of browsing times for each content, calculate the sum of each browsing content of the user multiplied by the corresponding browsing times, and obtain the preference value V of the user; Obtain the feature vectors of the content to be pushed for each type; According to the formula , calculate and obtain the preference weight of the user for each category , where represents the preference weight for the th category, is the total number of content categories, is the preference value of this user for the th category of content, where is any integer from 1 to , representing different content categories; According to the formula , the content score of each content is obtained , where is the value of each type of content on the th category, represents the preference weight of the th category; Obtain the freshness B of the content; According to the formula , the final recommendation score is obtained . Set a threshold for the recommendation score, and transmit the push content with a recommendation score higher than the threshold to the push unit.

8. An item planning and business intelligence push system based on AI and knowledge graph according to claim 7, characterized in that, The specific working mode of the push unit is as follows: Enrich the recommendation list of the user with the content with a high recommendation score, recommend it to the user, and at the same time collect the feedback and engagement of the user on the pushed content.

Images