Financial information recommendation method and system based on multi-channel distribution

Abstract

The invention belongs to the technical field of information recommendation, and discloses a financial information recommendation method and system based on multi-channel distribution, and the method comprises the steps: collecting behavior data of multiple channels of a user, carrying out the cross-platform data integration, and forming a complete user behavior data set; constructing a multi-dimensional user portrait, establishing a financial information database, and marking a keyword tag for each piece of financial information; according to the keyword matching degree of the user portrait and the information content, using content recommendation and financial information conforming to user preferences, and based on historical behavior data of the user and behavior habits of similar users, using collaborative filtering to recommend financial information in which the user is interested to obtain a financial information recommendation list; according to the use habits of the user at different terminals, cross-platform information distribution is performed on the financial information recommendation list through multiple channels, so that the accurate recommendation capability of financial information is improved, and the user experience is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of information recommendation, and in particular relates to a financial information recommendation method and system based on multi-channel distribution. Background Art

[0002] With the rapid development of the Internet, the amount and speed of financial information release have exploded. Various financial information platforms, social media, news portals, etc. continue to push a large amount of financial dynamics and analysis reports. However, when faced with such a large amount of information, users often find it difficult to find the information content that they are really interested in and valuable. This information overload phenomenon leads to distracted users and low efficiency in obtaining information.

[0003] Most of the current financial information distribution systems on the market have a single channel distribution, insufficient user coverage, incomplete user behavior data, and difficulty in building accurate user portraits. Many information platforms rely only on a single channel for information push, such as web pages or mobile apps. Since users are accustomed to obtaining information on different terminals and platforms, the single-channel distribution method cannot adapt to the diverse needs of users. Most existing recommendation systems rely only on user behavior data on a single platform (such as only considering the user's browsing behavior on the APP), and it is difficult to integrate the user's behavior trajectories on multiple platforms. This fragmented data situation leads to an incomplete user portrait construction, and it is impossible to accurately capture the user's behavior patterns and interest changes in different scenarios and time periods. Since the user's behavior data on different platforms cannot be effectively integrated, the system cannot build a comprehensive portrait of the user, resulting in limited accuracy of personalized recommendations. Even for large information platforms, it is difficult to achieve true cross-platform data integration and behavior analysis. The needs of users in the financial market are very dynamic. Affected by market fluctuations, policy changes, economic events, etc., the information needs of users in different periods will change rapidly, and the existing recommendation system cannot perceive the changes in user needs in real time, resulting in the recommendation content lagging behind the user's needs.

[0004] Therefore, in order to solve the problem of how to build a comprehensive user portrait and realize accurate recommendation and multi-channel distribution of financial information across platforms to meet the personalized needs of users on different platforms and terminals, a multi-channel distribution financial information recommendation method and system is urgently needed, which aims to improve the accurate recommendation ability of financial information and enhance user experience through collaborative distribution across multiple platforms and multiple devices. Summary of the invention

[0005] To solve the above problems, the present invention provides a financial information recommendation method and system based on multi-channel distribution, which can combine the user's behavioral data on different platforms to construct a multi-dimensional user portrait, and make accurate information recommendations based on this, thereby achieving a seamless information experience for users on multiple channels and multiple terminals.

[0006] The financial information recommendation method based on multi-channel distribution described in the present invention comprises the following steps: S10. Collect user behavior data from multiple channels through the front-end system of the financial information platform; S20, integrating the user's behavior data across platforms to form a complete user behavior data set; S30. Analyze the content of financial information based on the collected user behavior data set, extract topic tags in the financial field, and build a multi-dimensional user portrait; S40. Classify the financial information content on the platform, establish a financial information database, and tag each piece of financial information with keywords; S50, using the user portrait and the keyword matching degree of the information content, using the content to recommend financial information that matches the user's preferences, and based on the user's historical behavior data and the behavior habits of user groups with similar interests and behavior patterns, using a collaborative filtering algorithm to recommend financial information that the user is interested in, and obtaining a financial information recommendation list; S60. Distribute the financial information recommendation list across platforms through multiple channels based on the user's usage habits on different terminals.

[0007] As a further solution of the present invention, in S10, the user's behavior data on multiple channels is collected, including web pages, mobile terminals and social platforms, wherein the web pages and mobile terminals are access platforms, and the social platforms are interactive platforms, and the collected behavior data include browsing history, number of clicks, length of stay, likes, comments and sharing behavior data.

[0008] As a further solution of the present invention, S20 is specifically: Use platform login information to match user identity; if the user is not logged in, generate a temporary ID based on device information for data collection; if the user changes the platform login, merge the temporary ID with the global user ID, and associate the temporary login behavior data with the global ID; Determine the fields and formats of user behavior data on each platform and establish a standardized behavior data model; Convert the format of the collected data and store it according to the standardized behavior data model to remove duplicate behavior records; The deduplicated behavior data is integrated into a complete user behavior data set and stored uniformly in a NoSQL database. During storage, a data storage entity is created for each user, which contains all the user's cross-platform behavior data, and the behavior data is automatically appended to the user's behavior data set through the global user ID.

[0009] As a further solution of the present invention, in S30, analyzing the content of financial information based on the collected user behavior data set includes the following steps: Perform word segmentation and stop word removal preprocessing on the collected financial information text to obtain text data with word form restoration and punctuation removal; Use TF-IDF to identify keywords in financial information that appear more frequently than a preset ratio and are distinguishable in different information texts; The LDA topic model is used to extract topics from the text data of financial information and assign multiple topic tags to each information text.

[0010] As a further solution of the present invention, when constructing a multi-dimensional user portrait, the user's interest preferences, active time periods, and reading habits are analyzed based on the collected user behavior data set, and the financial information content read by the user is extracted and marked with topic tags in the financial field. The topic tags include stocks, funds, and bonds.

[0011] As a further solution of the present invention, constructing a multi-dimensional user portrait includes the following steps: Through the collected user behavior data set, statistics are collected on each user's interaction with financial information; Generate an interest model for each user, the interest model calculating the user's preference for a certain type of financial information based on the frequency and time dimension of the information topic tags read by the user; Based on the constructed user profile, collaborative filtering algorithms are used to recommend information that best matches the user's interests; Store each user's profile in the database for subsequent recommendation system calls.

[0012] As a further solution of the present invention, in S40, when a financial information database is established, the financial information database is a structured database including the following tables: Information table: contains information ID, title, content, release time, category label, and keyword label; Category table: contains category ID, category name, and description; Keyword table: contains keyword ID, keyword name, and related information ID.

[0013] As a further solution of the present invention, when tagging each piece of financial information with keywords, the RAKE algorithm is used to extract keywords. The specific steps include: segmenting the text and removing stop words, extracting candidate keywords, and obtaining the score of each candidate word by calculating the word frequency, word co-occurrence and word weight, and selecting the keyword with the highest score as the final label; the score is calculated as follows: score = (word frequency × weight) × (co-occurrence frequency).

[0014] As a further solution of the present invention, S50 is specifically: Collect users' basic information, interest preferences, and historical behavior data, extract users' interest characteristics and behavior characteristics, and form user portraits; Extract keywords and themes from financial information content to form information feature vectors, and classify information content by theme; By calculating the matching degree between the interest keywords in the user portrait and the keywords in the information content, the cosine similarity is used to quantify it, and a matching score is generated for each piece of information; Build a user behavior matrix to record the user's interaction data on information content, use user collaborative filtering to identify users with similar behaviors to the current user, and recommend information that the current user has not been exposed to based on the behaviors of similar users; The keyword matching degree and collaborative filtering recommendation results are weighted and integrated to generate the final recommendation list.

[0015] As a further solution of the present invention, in S60, the financial information recommendation list is distributed across platforms through multiple channels, including the following steps: adjusting the display format of the recommended content according to the terminal characteristics, performing terminal adaptation, and matching the push channel according to the recommended content, and distributing the recommended financial information across platforms to the user terminal, wherein the terminal characteristics include screen size and operating system.

[0016] The present invention also provides a financial information recommendation system based on multi-channel distribution, comprising: The data collection module is used to collect user behavior data from multiple channels through the front-end system of the financial information platform; Data integration module, used to integrate user behavior data across platforms to form a complete user behavior data set; The portrait generation module is used to analyze the content of financial information based on the collected user behavior data set, extract the topic tags in the financial field, and build a multi-dimensional user portrait; The keyword matching module is used to classify the financial information content on the platform, establish a financial information database, and tag each piece of financial information with a keyword; The information recommendation module is used to recommend financial information that matches the user's preferences by using the matching degree between the user's profile and the keywords in the information content, and to recommend financial information that the user is interested in by using collaborative filtering based on the user's historical behavior data and the behavior habits of user groups with similar interests and behavior patterns, thereby obtaining a list of recommended financial information; The cross-platform distribution module is used to distribute the financial information recommendation list across platforms through multiple channels based on the user's usage habits on different terminals.

[0017] As a further solution of the present invention, the data acquisition module includes a front-end system integration unit and a data acquisition interface. The front-end system integration unit is responsible for collecting user behavior data, such as browsing, clicking, dwell time, likes, comments, etc., from multiple terminals of the financial information platform (including web pages, mobile terminals, APPs, mini-programs, etc.); the data acquisition interface is used to collect user cross-terminal behavior data through an API interface or SDK embedding method to ensure the comprehensiveness and consistency of data collection.

[0018] As a further solution of the present invention, the data integration module includes a data processing engine and a user identity management unit. The data processing engine is used to deduplicate, clean and normalize user data from different terminals to form a complete user behavior data set. The user identity management unit is used to uniformly identify the user's behavior on different terminals through user login information, device ID, Cookie, etc., to ensure the accuracy and consistency of cross-platform data.

[0019] As a further solution of the present invention, the portrait generation module includes a financial information analysis unit, which includes an information content analysis tool and an information database. The information content analysis tool is used to perform text analysis on financial information, extract the subject, keywords and other important information of the information, and generate subject tags (such as stocks, funds, bonds, money markets, etc.); the information database is used to build a financial information database to store all classified and labeled financial information to ensure that the system can access and match the latest information content at any time.

[0020] As a further solution of the present invention, the portrait generation module includes a user behavior feature extractor and a portrait update engine. The user behavior feature extractor is used to extract the user's interest preferences (such as financial fields of interest, investment style, browsing time period, etc.) and usage habits (such as preferred terminals, reading frequency, etc.) based on the user behavior data set to construct a multi-dimensional user portrait; the portrait update engine is used to dynamically update the user portrait based on the user's latest behavior data to ensure the timeliness and accuracy of the portrait.

[0021] As a further solution of the present invention, the information recommendation module includes a keyword matching recommendation engine, a collaborative filtering recommendation engine and a hybrid recommendation unit. The keyword matching recommendation engine is used to generate a recommended financial information list by calculating the similarity between the interest keywords in the user portrait and the financial information content keywords; the collaborative filtering recommendation engine is used to recommend financial information that the user may be interested in but has not been exposed to based on the user's historical behavior data and the behavior habits of similar users, using a collaborative filtering algorithm; the hybrid recommendation unit is used to combine keyword matching and collaborative filtering to comprehensively generate a personalized information recommendation list to ensure the diversity and relevance of the recommended content.

[0022] As a further solution of the present invention, the platform distribution module includes a channel adapter, a cross-platform push unit and a real-time feedback collector. The channel adapter is used to automatically adjust the presentation form of the recommended content according to the characteristics of the user terminal (such as screen size, operating system, network environment, etc.) to ensure the best display effect; the cross-platform push unit is used to push the recommended financial information to the user through channels such as message push, SMS, email, social media, etc., to achieve cross-platform information distribution; the real-time feedback collector is used to collect user interactive feedback (such as click-through rate, unsubscribe, likes, etc.) through various channels for subsequent recommendation algorithm optimization.

[0023] As a further solution of the present invention, the financial information recommendation system based on multi-channel distribution also includes a data security and privacy management module, which includes: Data encryption engine: Encrypts user behavior data and sensitive information for storage and transmission to ensure data security; Privacy protection system: Comply with relevant privacy protection laws and regulations, such as GDPR, to ensure the legal collection and use of user privacy information, and provide privacy setting management functions to allow users to control data usage permissions.

[0024] As a further solution of the present invention, the financial information recommendation system based on multi-channel distribution also includes a recommendation algorithm optimization module, which includes: Feedback loop system: Based on user behavior feedback data, regularly evaluate the effectiveness of recommendations, adjust recommendation strategies, and optimize algorithms; A / B testing system: supports parallel testing of different recommendation strategies and automatically selects the most effective recommendation plan based on user feedback data.

[0025] Compared with the prior art, the financial information recommendation method and system based on multi-channel distribution proposed by the present invention has the following beneficial effects: 1. Accurately build user portraits; the present invention can conduct a comprehensive and detailed analysis of user behavior through cross-terminal and multi-channel data collection and integration. By deeply mining the usage habits of different terminals and combining historical behavior data, the system can dynamically and accurately build user portraits; this makes the system more accurate in predicting user interests and needs, ensuring that the recommended content is highly consistent with user preferences; 2. Realizes multi-dimensional personalized recommendations; the present invention combines keyword matching and collaborative filtering to recommend financial information based on the user's explicit interests (such as browsing content, keywords), and to provide personalized content recommendations by analyzing the behavioral habits of similar users; the hybrid recommendation algorithm ensures the diversity and relevance of the recommended content, and users can obtain information that is both in line with their interests and has a certain degree of freshness; 3. Realize cross-platform and multi-terminal information distribution; The present invention has a powerful cross-platform information distribution capability. Regardless of whether the user is using a mobile terminal, a desktop terminal or other intelligent terminal, the system can adaptively adjust the content display format according to the characteristics of the user terminal; through message push, email notification, social media push and other methods, the system can ensure that financial information reaches the user in a timely and accurate manner in the scenario where the user uses different terminals, greatly improving the smoothness of the user experience; 4. Real-time feedback and recommendation optimization are achieved; the present invention has the function of collecting user feedback (such as clicks, interactions, subscriptions and unsubscriptions, etc.) in real time, and can continuously optimize the recommendation algorithm based on the feedback data; through A / B testing, user behavior analysis and other means, the system can dynamically adjust the recommendation strategy to ensure that the recommended content keeps pace with user needs; user feedback forms a closed loop to ensure the continuous evolution of the recommendation system, and further improve the accuracy of recommendations and user satisfaction; 5. Data security and privacy protection are achieved; the present invention encrypts, stores and transmits user behavior data, strictly abides by GDPR and other data privacy protection laws and regulations, and ensures that user privacy information will not be abused; users can also choose data usage permissions independently, which increases the transparency of the system and the trust of users; through legal and compliant data usage strategies, users can not only enjoy personalized services, but also have no worries about privacy issues; 6. Improved user engagement and retention rate; the precise recommendation mechanism and cross-channel push capabilities of the present invention can effectively improve user stickiness and engagement; by providing financial information that is highly consistent with user interests, users stay on the platform longer, and interaction and conversion rates are improved; in addition, the timely and intelligent push mechanism can also attract user return, improve the platform's activity and retention rate, and thus increase user loyalty.

[0026] In summary, the present invention provides an efficient, intelligent, and personalized financial information recommendation method and system by integrating user behavior data, building accurate user portraits, achieving cross-platform distribution, multi-dimensional recommendations, and ensuring data security and privacy protection. The system not only greatly improves the user experience, but also continuously improves the recommendation effect through real-time optimization and data analysis, and promotes user growth and service optimization of the financial information platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flowchart of a financial information recommendation method based on multi-channel distribution according to an embodiment of the present invention; Figure 2 A flowchart of cross-platform data integration in a financial information recommendation method based on multi-channel distribution in an embodiment of the present invention; Figure 3 A flowchart of constructing a multi-dimensional user portrait in a financial information recommendation method based on multi-channel distribution in an embodiment of the present invention; Figure 4 A flowchart of obtaining a financial information recommendation list in a financial information recommendation method based on multi-channel distribution in an embodiment of the present invention; Figure 5 The following is a schematic diagram of the user portrait accuracy test results, where (a) is the consistency ratio comparison, (b) is the user click rate comparison, (c) is the user stay time comparison, and (d) is the user feedback ratio comparison; Figure 6 Schematic diagram of the test results of the recommendation system, where (a) is the comparison of click-through rates in different time periods, (b) is the comparison of conversion rates in different time periods, (c) is the change in user satisfaction scores, and (d) is the comparison of key indicators of A / B testing; Figure 7 Schematic diagram of the test results of cross-platform information distribution efficiency, where (a) is the comparison of information distribution time, (b) is the delay analysis of the distribution process, (c) is the user satisfaction score, and (d) is the distribution of user experience satisfaction scores. DETAILED DESCRIPTION

[0028] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments in conjunction with the accompanying drawings.

[0029] The existing information recommendation system lacks the ability to collaboratively distribute financial information across multiple channels, resulting in low efficiency in user information acquisition and difficulty in forming accurate and personalized recommendation effects. The present invention proposes a financial information recommendation method and system based on multi-channel distribution, which can combine user behavior data on different platforms to build a multi-dimensional user portrait, and based on this, make accurate information recommendations, thereby achieving a seamless information experience for users across multiple channels and multiple terminals.

[0030] like Figure 1 As shown, an embodiment of the present invention provides a financial information recommendation method based on multi-channel distribution, the method comprising the following steps:

[0031] S10. Collect user behavior data from multiple channels through the front-end system of the financial information platform.

[0032] In this step, the behavioral habits of user groups with similar interests and behavioral patterns are analyzed based on the user's historical behavior data.

[0033] In the data sources of this embodiment, user behavior data comes from different financial information platforms, including but not limited to: mobile apps, desktop browsers, social media, and mini-programs. The types of user behavior data collected by financial information platforms include: Browsing history (e.g. articles read, information clicked); Search history (keywords that users actively search for); Interaction data (likes, comments, shares, etc.); Device information (device type, operating system, screen size, etc.).

[0034] When cross-platform association is performed, a unique user ID is generated for each user, and data from different channels is matched with the user ID; when deduplicating and verifying data, duplicate data across platforms is ensured to be deduplicated, and data integrity and accuracy are verified. When unifying the data storage format, the data formats of different platforms are different, so format standardization is required for subsequent data processing and analysis.

[0035] S20. Integrate the user's behavior data across platforms to form complete user behavior data.

[0036] In this step, if Figure 2 As shown, the user's behavior data is integrated across platforms, including the following steps: S201. Use platform login information to match user identity; if the user is not logged in, generate a temporary ID based on device information for data collection; if the user changes the platform login, merge the temporary ID with the global user ID, and associate the temporary login behavior data with the global ID; S202, determine the fields and formats of user behavior data on each platform, and establish a standardized behavior data model; S203, converting the format of the collected data and storing it according to a standardized behavior data model, removing duplicate behavior records; S204, integrating the deduplicated behavior data into a complete user behavior data set, and using a NoSQL database for unified storage, wherein, during storage, a data storage entity is created for each user, which contains all the user's cross-platform behavior data, and the behavior data is automatically appended to the user's behavior data set through the global user ID.

[0037] In this embodiment, in order to ensure that the behavior data of the same user on multiple platforms can be associated, a globally unique user identifier (user ID mapping) needs to be generated for each user. The identifier is achieved through the following steps: 1) Use login information from different platforms (such as user mobile phone number, email address, social account ID, etc.) to match user identities; 2) If the user is not logged in, a temporary ID can be generated based on the device information for data collection; the device information can include IP address, device type, screen resolution, operating system, etc., and a specific rule combination is used to generate the device ID; 3) If the user subsequently logs in on another platform, the temporary ID is merged with the global user ID, and the temporary login behavior data is associated with the global ID; When standardizing and deduplicating data, the data formats of different platforms are usually different, so the data needs to be standardized and duplicate data needs to be removed before integration. For financial information platforms with high real-time requirements, user behavior data needs to be synchronized in real time between multiple platforms. Use message queues (such as Kafka) to stream user behavior data to the background processing module to ensure that the data can be quickly integrated and updated into the user data set.

[0038] S30. Analyze the financial information content based on the collected user behavior data set, extract the topic tags in the financial field, and build a multi-dimensional user portrait.

[0039] In this step, when analyzing the financial information content based on the collected user behavior data set, specifically: Perform word segmentation and stop word removal preprocessing on the collected financial information text to obtain text data with word form restoration and punctuation removal; Use TF-IDF to identify keywords that appear frequently in financial information and are distinguishable in different information texts; The LDA topic model is used to extract topics from the text data of financial information and assign multiple topic tags to each information text.

[0040] Among them, the extraction of topics from the text data of financial information using the LDA topic model is achieved through the following steps: Initialize the topic distribution in the document and set the number of topics K; Iteratively update the topic assignment of each word based on the co-occurrence relationship between documents and topics; Output the topic distribution of each financial information, such as: Document 1 topic distribution = [topic 1: 0.4, topic 2: 0.3, topic 3: 0.3]; Examples of topic tags: stock market, monetary policy, macroeconomics, etc.

[0041] In this embodiment, when constructing a multi-dimensional user portrait, the user's interest preferences, active time periods, and reading habits are analyzed based on the collected user behavior data set, and the financial information content that the user has read is extracted and labeled with topic tags in the financial field. The topic tags include stocks, funds, and bonds.

[0042] Among them, user portrait is the modeling of user behavior and interests. When building user portrait, you can also analyze it from the following dimensions: User preference dimension: the types of information or topic tags that users read, click, save, and share; User behavior dimension: user visit frequency, stay time, interaction (like, comment, forward); User device dimension: the type of device, operating system, browser, etc. used by the user; User time and space dimension: user's visit time period, visit location, etc.; When constructing a user portrait, the collected user behavior data set is used to count each user's interaction with financial information, for example: number of article clicks: the number of articles the user has read; frequency of clicks on topic tags: the topic tags that users read most, such as "stock market" and "macroeconomics".

[0043] Among them, Figure 3 As shown in the figure, building a multi-dimensional user portrait includes the following steps: S301. Counting the interaction between each user and financial information through the collected user behavior data set; S302, generating an interest model for each user, the model calculating the user's preference for a certain type of financial information based on the frequency and time dimension of the information topic tags read by the user; S303, based on the constructed user portrait, using collaborative filtering algorithm to recommend information that best matches the user's interests; S304, storing the portrait of each user in a database for subsequent recommendation system calls; Among them, user-based collaborative filtering: recommend information that similar users have read; item-based collaborative filtering: recommend content similar to the information the user has read, and store the portrait of each user in the database for subsequent recommendation system calls.

[0044] S40. Classify the financial information content on the platform, establish a financial information database, and tag each piece of financial information with keywords.

[0045] In this step, when establishing a financial information database, the financial information database is a structured database, which includes the following tables: Information table: contains information ID, title, content, release time, category label, and keyword label; Category table: contains category ID, category name, and description; Keyword table: contains keyword ID, keyword name, and related information ID.

[0046] In this embodiment, when a keyword tag is added to each piece of financial information, the RAKE algorithm is used to extract keywords. The specific steps include: segmenting the text and removing stop words, extracting candidate keywords, and obtaining a score for each candidate word by calculating the word frequency, word co-occurrence, and word weight. The keyword with the highest score is selected as the final tag. The score is calculated as follows: score = (word frequency × weight) × (co-occurrence frequency).

[0047] S50. Use content recommendations based on the keyword matching between user portraits and information content to obtain financial information that matches the user's preferences. Based on the user's historical behavior data and the behavior habits of user groups with similar interests and behavior patterns, use a collaborative filtering algorithm to recommend financial information that the user is interested in, and obtain a list of recommended financial information.

[0048] In this step, if Figure 4 As shown, obtaining a list of recommended financial information includes the following steps: S501, collecting the user's basic information, interest preferences and historical behavior data, extracting the user's interest characteristics and behavior characteristics, and forming a user portrait; S502, extracting keywords and themes from the financial information content to form an information feature vector, and classifying the information content according to themes; S503, by calculating the matching degree between the interest keywords in the user portrait and the information content keywords, quantifying them using cosine similarity, and generating a matching score for each piece of information; S504, constructing a user behavior matrix, recording the user's interactive data on information content, using user collaborative filtering to identify users with similar behaviors to the current user, and recommending information that the current user has not been exposed to based on the behaviors of similar users; S505: Perform weighted fusion on the keyword matching degree and the collaborative filtering recommendation result to generate a final recommendation list.

[0049] S60, distributing the financial information recommendation list across platforms through multiple channels according to the user's usage habits on different terminals, including the following steps: The display format of recommended content is adjusted according to terminal characteristics, terminal adaptation is performed, and push channels are matched according to recommended content to distribute recommended financial information to user terminals across platforms. Terminal characteristics include screen size and operating system.

[0050] The financial information recommendation method based on multi-channel distribution of the present invention can conduct a comprehensive and detailed analysis of user behavior through cross-terminal and multi-channel data collection and integration. By deeply mining the usage habits of different terminals and combining historical behavior data, the system can dynamically and accurately build user portraits. This makes the system more accurate in predicting user interests and needs, ensuring that the recommended content is highly consistent with the user's preferences.

[0051] The present invention combines keyword matching and collaborative filtering to recommend financial information based on the user's explicit interests (such as browsing content, keywords), and to provide personalized content recommendations by analyzing the behavioral habits of similar users. The hybrid recommendation algorithm ensures the diversity and relevance of the recommended content, and users can obtain information that not only meets their interests but also has a certain sense of freshness; the present invention has a powerful cross-platform information distribution capability. Regardless of whether the user is using a mobile terminal, desktop terminal or other smart terminal, the system can adaptively adjust the content display format according to the characteristics of the user terminal. Through message push, email notification, social media push and other methods, the system can ensure that financial information reaches the user in a timely and accurate manner in scenarios where the user uses different terminals, greatly improving the smoothness of the user experience.

[0052] The present invention has the function of collecting user feedback (such as clicks, interactions, subscriptions and unsubscriptions) in real time, and can continuously optimize the recommendation algorithm based on the feedback data. Through means such as A / B testing and user behavior analysis, the system can dynamically adjust the recommendation strategy to ensure that the recommended content keeps pace with user needs. User feedback forms a closed loop to ensure the continuous evolution of the recommendation system and further improve the accuracy of the recommendation and user satisfaction. The present invention encrypts, stores and transmits user behavior data, strictly abides by data privacy protection laws and regulations such as GDPR, and ensures that user privacy information will not be abused. Users can also choose data usage rights independently, which increases the transparency of the system and the trust of users. Through legal and compliant data usage strategies, users can not only enjoy personalized services, but also have no worries about privacy issues. The precise recommendation mechanism and cross-channel push capabilities of the present invention can effectively improve the user's stickiness and participation. By providing financial information that is highly consistent with user interests, users' stay time on the platform increases, and the interaction and conversion rate are improved. In addition, the timely and intelligent push mechanism can also attract user return, improve the platform's activity and retention rate, and thus increase user loyalty.

[0053] In a second aspect of the embodiments of the present invention, the present invention further provides a financial information recommendation system based on multi-channel distribution, comprising: The data collection module is used to collect user behavior data from multiple channels through the front-end system of the financial information platform; Data integration module, used to integrate user behavior data across platforms to form a complete user behavior data set; The portrait generation module is used to analyze the content of financial information based on the collected user behavior data set, extract the topic tags in the financial field, and build a multi-dimensional user portrait; The keyword matching module is used to classify the financial information content on the platform, establish a financial information database, and tag each piece of financial information with a keyword; The information recommendation module is used to recommend financial information that matches the user's preferences based on the keyword matching between the user's profile and the information content, and to recommend financial information that the user is interested in based on the user's historical behavior data and the behavior habits of user groups with similar interests and behavior patterns using collaborative filtering to obtain a list of recommended financial information; The cross-platform distribution module is used to distribute the financial information recommendation list across platforms through multiple channels based on the user's usage habits on different terminals.

[0054] Among them, the data collection module includes a front-end system integration unit and a data collection interface. The front-end system integration unit is responsible for collecting user behavior data, such as browsing, clicking, dwell time, likes, comments, etc., from multiple terminals of the financial information platform (including web pages, mobile terminals, APPs, mini-programs, etc.); the data collection interface is used to collect user cross-terminal behavior data through an API interface or SDK embedding method to ensure the comprehensiveness and consistency of data collection.

[0055] In this embodiment, the data integration module includes a data processing engine and a user identity management unit. The data processing engine is used to deduplicate, clean and normalize user data from different terminals to form a complete user behavior data set. The user identity management unit is used to uniformly identify the user's behavior on different terminals through user login information, device ID, Cookie, etc., to ensure the accuracy and consistency of cross-platform data.

[0056] In this embodiment, the portrait generation module includes a financial information analysis unit, which includes an information content analysis tool and an information database. The information content analysis tool is used to perform text analysis on financial information, extract the subject, keywords and other important information of the information, and generate subject tags (such as stocks, funds, bonds, money markets, etc.); the information database is used to build a financial information database to store all classified and labeled financial information to ensure that the system can access and match the latest information content at any time.

[0057] In this embodiment, the portrait generation module includes a user behavior feature extractor and a portrait update engine. The user behavior feature extractor is used to extract the user's interest preferences (such as financial fields of interest, investment style, browsing time period, etc.) and usage habits (such as preferred terminals, reading frequency, etc.) based on the user behavior data set to construct a multi-dimensional user portrait; the portrait update engine is used to dynamically update the user portrait based on the user's latest behavior data to ensure the timeliness and accuracy of the portrait.

[0058] In this embodiment, the information recommendation module includes a keyword matching recommendation engine, a collaborative filtering recommendation engine and a hybrid recommendation unit. The keyword matching recommendation engine is used to generate a recommended financial information list by calculating the similarity between the interest keywords in the user portrait and the financial information content keywords; the collaborative filtering recommendation engine is used to recommend financial information that the user may be interested in but has not been exposed to based on the user's historical behavior data and the behavior habits of similar users, using a collaborative filtering algorithm; the hybrid recommendation unit is used to combine keyword matching and collaborative filtering to comprehensively generate a personalized information recommendation list to ensure the diversity and relevance of the recommended content.

[0059] In this embodiment, the platform distribution module includes a channel adapter, a cross-platform push unit and a real-time feedback collector. The channel adapter is used to automatically adjust the presentation form of the recommended content according to the characteristics of the user terminal (such as screen size, operating system, network environment, etc.) to ensure the best display effect; the cross-platform push unit is used to push the recommended financial information to the user through channels such as message push, SMS, email, social media, etc., to achieve cross-platform information distribution; the real-time feedback collector is used to collect user interactive feedback (such as click-through rate, unsubscribe, likes, etc.) through various channels for subsequent recommendation algorithm optimization.

[0060] In this embodiment, the financial information recommendation system based on multi-channel distribution also includes a data security and privacy management module, which includes: Data encryption engine: Encrypts user behavior data and sensitive information for storage and transmission to ensure data security; Privacy protection system: Comply with relevant privacy protection laws and regulations, such as GDPR, to ensure the legal collection and use of user privacy information, and provide privacy setting management functions to allow users to control data usage permissions.

[0061] In this embodiment, the financial information recommendation system based on multi-channel distribution further includes a recommendation algorithm optimization module, which includes: Feedback loop system: Based on user behavior feedback data, regularly evaluate the effectiveness of recommendations, adjust recommendation strategies, and optimize algorithms; A / B testing system: supports parallel testing of different recommendation strategies and automatically selects the most effective recommendation plan based on user feedback data.

[0062] Through the above detailed steps and examples, the financial information recommendation system based on multi-channel distribution of the present invention is used to execute the steps of the financial information recommendation method based on multi-channel distribution in the above embodiment, which will not be repeated here. The financial information recommendation system based on multi-channel distribution of the present invention provides an efficient, intelligent and personalized financial information recommendation method and system by integrating user behavior data, building accurate user portraits, realizing cross-platform distribution, multi-dimensional recommendation, and ensuring data security and privacy protection. The system not only greatly improves the user experience, but also continuously improves the recommendation effect through real-time optimization and data analysis, and promotes user growth and service optimization of the financial information platform.

[0063] In order to verify the performance of the technical solution of the present invention, the following experiments were conducted.

[0064] (1) User portrait accuracy test;

[0065] Experimental background: In the field of information recommendation, the accuracy of user portraits is one of the key indicators to measure the performance of recommendation systems. An accurate user portrait can ensure that the recommended content is highly matched with the user's actual interests and needs, thereby improving user experience and satisfaction. This experiment aims to evaluate the accuracy of the method in constructing user portraits, that is, to evaluate the consistency between the user portraits generated by the system and the actual behavior of the user.

[0066] Experimental method: Collect the interactive data of users participating in the experiment on financial information over a period of time, including clicks, browsing, stay time, likes, comments and sharing; at the same time, collect the user's demographic information, such as age, gender, occupation and educational background, as auxiliary data for user portraits. User portrait construction: using the method of the present invention, combined with the user's behavioral data and demographic information, to construct a multi-dimensional user portrait for each user; the user portrait includes key features such as the user's interest preferences, active time period, reading habits, and is marked with topic tags in the financial field, such as stocks, funds and bonds. Feedback collection and analysis: for the information recommended by the system, collect user feedback data, including click-through rate, reading time and interactive behavior; analyze the consistency between user feedback and user portraits, and evaluate the accuracy of user portraits. Consistency evaluation: design a consistency evaluation model, compare the user's actual behavior with the interest preferences in the user portrait, and calculate the consistency ratio; use machine learning algorithms, such as random forests or gradient boosting machines, to quantitatively evaluate the accuracy of user portraits.

[0067] Result analysis: The user portrait accuracy test results are as follows Figure 5 As shown, Figure 5(a) is the consistency ratio comparison, Figure 5 (b) is the comparison of user click rate. Figure 5 (c) is the comparison of user stay time. Figure 5 (d) is a comparison of user feedback ratios. The consistency between the user portrait constructed by the method of the present invention and the actual user behavior reaches more than 90%, which is much higher than the 75% of the traditional method; the accuracy of the method of the present invention in identifying user interest preferences reaches 92%, the accuracy in predicting active time periods reaches 88%, and the accuracy in matching reading habits reaches 91%. Compared with the traditional method, the user group using the method of the present invention has a 30% higher click-through rate, a 20% higher reading time, and a 40% higher interactive behavior; user feedback shows that the method of the present invention can more accurately capture the user's interests and provide information recommendations that better meet user needs. Case analysis shows that for specific user groups, such as young investors, the method of the present invention can more accurately identify their interest in emerging markets and high-tech stocks, and the relevance of recommendations is increased by 50% compared with traditional methods; for experienced investors, the method of the present invention can better understand their needs for market analysis and macroeconomic information, and the relevance of recommendations is increased by 45%. Algorithm optimization: Based on the experimental results, the recommendation algorithm was optimized, especially in the real-time processing of user behavior data and the dynamic updating of user portraits; the optimized algorithm can respond to changes in user behavior more quickly and improve the timeliness and accuracy of user portraits.

[0068] (2) Recommendation system effectiveness testing;

[0069] Experimental background: In the field of financial information, the accuracy of the recommendation system is crucial to improving user experience and enhancing user stickiness. An accurate recommendation system can push information that users are truly interested in and valuable, thereby improving users' click-through rate, conversion rate and satisfaction. This experiment aims to evaluate the accuracy of the method of the present invention in recommending financial information and compare it with the existing traditional recommendation system. The purpose of this experiment is to verify the effectiveness of the recommendation system of the present invention in recommending financial information through comparative experiments. Through A / B testing, we will evaluate and compare the differences between the recommendation system of the present invention and the traditional recommendation system in key performance indicators, including click-through rate, conversion rate and user satisfaction.

[0070] Experimental method: Design an A / B test and randomly divide users into two groups with an equal number of users in each group to ensure the fairness of the experiment; one group uses the recommendation system of the present invention (Group A), and the other group uses the traditional recommendation system (Group B). Data collection: collect interactive data of the two groups of users, including click-through rate, reading time, conversion rate (such as registration, purchase, etc.) and user feedback; click-through rate is defined as the ratio of the number of times the recommended information is clicked to the number of times it is recommended; conversion rate is defined as the ratio of the number of times the target behavior (such as registration, purchase, etc.) occurs to the number of times the recommended information is clicked. User satisfaction survey: collect user satisfaction evaluations on recommended information through online questionnaires or in-app feedback mechanisms. User satisfaction is quantified by scoring (1-5 points), where 5 points are very satisfied and 1 point is very dissatisfied.

[0071] Result analysis: Recommendation system effect test Figure 6 As shown, Figure 6 (a) is the comparison of click-through rates in different time periods. Figure 6 (b) is the comparison of conversion rates in different time periods. Figure 6 (c) is the change in user satisfaction score, Figure 6 (d) is the comparison of key indicators of A / B test. The average click-through rate of users in group A is 30% higher than that of users in group B, indicating that the recommendation system of the present invention can more accurately capture user interests and improve users' willingness to click on recommended information; analysis shows that the recommendation system of the present invention can more effectively push information related to user interests through keyword matching and collaborative filtering algorithms. Conversion rate analysis shows that the conversion rate of users in group A is 30% higher than that of users in group B, indicating that the recommendation system of the present invention can more effectively promote users' target behaviors; this result is attributed to the accuracy of the recommendation system of the present invention in user portrait construction and information matching, which makes the recommended content more in line with users' actual needs. User satisfaction analysis shows that the average satisfaction score of users in group A is 40% higher than that of users in group B, showing the high satisfaction of users with the recommendation system of the present invention; user feedback shows that the information provided by the recommendation system of the present invention is more personalized and relevant, which improves the reading experience of users. Case study: For specific user groups, such as novice investors and experienced investors, the recommendation system of the present invention shows higher accuracy and satisfaction; novice investors said that the recommendation system of the present invention helps them quickly find valuable market information, while experienced investors believe that the recommended content is more in-depth and professional.

[0072] (3) Cross-platform information distribution efficiency test;

[0073] Experimental background: In the digital age, users obtain information through a variety of terminals and platforms, and the efficiency of cross-platform information distribution is crucial to meet users' demand for real-time information. An efficient information distribution system can ensure that users can receive the latest financial information in a timely manner regardless of which platform they are on, thereby improving user experience and satisfaction. This experiment aims to evaluate the efficiency of the method of the present invention in cross-platform information distribution and determine its performance in practical applications. The purpose of this experiment is to evaluate the speed and efficiency of the method of the present invention in the process of cross-platform information distribution by simulating different user terminal usage scenarios. The experiment will record the time required for the entire process from information release to user reception, and analyze the factors affecting distribution efficiency.

[0074] Experimental method: simulate various user terminal usage scenarios, including web pages, mobile terminals (iOS and Android), social media platforms, etc.; select representative financial information as test content to ensure the consistency of test content. Data recording, record the entire process time from information release to user reception on each platform, including information release, server processing, network transmission and user reception; use timestamps to record the start and end time points of each link to ensure data accuracy. Efficiency evaluation, compare the difference in information distribution efficiency between the method of the present invention and the traditional distribution method; evaluate the delay in the information distribution process, including server processing delay, network transmission delay and terminal reception delay. Performance optimization, analyze experimental data, identify bottlenecks and problem points that affect information distribution efficiency; optimize the method of the present invention, including server configuration optimization, network transmission protocol optimization and terminal adaptation optimization.

[0075] Result analysis: The test results of cross-platform information distribution efficiency are as follows: Figure 7 As shown, Figure 7 (a) is the information distribution time comparison, Figure 7 (b) is the delay analysis of the distribution process. Figure 7 (c) is the user satisfaction score, Figure 7(d) is the distribution of user experience satisfaction scores. In terms of distribution time comparison, the average information distribution time of the method of the present invention in all test scenarios is 1 second, which is more than 50% shorter than that of traditional methods; on the web side, the average distribution time of the method of the present invention is 0.8 seconds, 1.1 seconds on the mobile side, and 0.9 seconds on the social media platform, showing good cross-platform consistency. Delay analysis, the server processing delay is reduced to 0.1 seconds, the network transmission delay is reduced to 0.3 seconds, and the terminal receiving delay is reduced to 0.2 seconds in the method of the present invention; by optimizing the server configuration and network transmission protocol, the method of the present invention significantly reduces the delay and improves the real-time nature of information distribution. Performance optimization effect, the optimized server configuration can process information release requests faster, the optimization of the network transmission protocol reduces the transmission time of data packets, and the terminal adaptation optimization ensures the rapid display of information on different devices; the system after performance optimization can still maintain a stable distribution efficiency under high load, meeting the needs of large-scale users. Long-term monitoring, establish a long-term monitoring mechanism, continuously track the efficiency of information distribution, and make dynamic adjustments based on user behavior and feedback; long-term monitoring results show that the method of the present invention can adapt to market changes and dynamic changes in user needs and maintain high efficiency of information distribution.

[0076] The above description is only a preferred embodiment of the present invention and is not intended to be a further limitation of the present invention. All equivalent changes made using the contents of the present specification and drawings are within the protection scope of the present invention.

Claims

1. A financial information recommendation method based on multi-channel distribution, characterized in that: The following steps are involved: S10. Collect user behavior data from multiple channels through the front-end system of the financial information platform; S20, integrating the user's behavior data across platforms to form a complete user behavior data set; S30. Analyze the content of financial information based on the collected user behavior data set, extract topic tags in the financial field, and build a multi-dimensional user portrait; S40. Classify the financial information content on the platform, establish a financial information database, and tag each piece of financial information with keywords; S50, using the user portrait and the keyword matching degree of the information content, using the content to recommend financial information that matches the user's preferences, and based on the user's historical behavior data and the behavior habits of user groups with similar interests and behavior patterns, using a collaborative filtering algorithm to recommend financial information that the user is interested in, and obtaining a financial information recommendation list; S60. Distribute the financial information recommendation list across platforms through multiple channels based on the user's usage habits on different terminals.

2. The method for recommending financial information based on multi-channel distribution according to claim 1, characterized in that: In S10, the user's behavior data on multiple channels is collected, including web pages, mobile terminals and social platforms, where the web pages and mobile terminals are access platforms and the social platforms are interactive platforms. The collected behavior data include browsing history, number of clicks, length of stay, likes, comments and sharing behavior data.

3. The financial information recommendation method based on multi-channel distribution according to claim 2, characterized in that: S20 is specifically: Use platform login information to match user identity; if the user is not logged in, generate a temporary ID based on device information for data collection; if the user changes the platform login, merge the temporary ID with the global user ID, and associate the temporary login behavior data with the global ID; Determine the fields and formats of user behavior data on each platform and establish a standardized behavior data model; Convert the format of the collected data and store it according to the standardized behavior data model to remove duplicate behavior records; The deduplicated behavior data is integrated into a complete user behavior data set and stored uniformly in a NoSQL database. During storage, a data storage entity is created for each user, which contains all the user's cross-platform behavior data, and the behavior data is automatically appended to the user's behavior data set through the global user ID.

4. The method for recommending financial information based on multi-channel distribution according to claim 3, characterized in that: In S30, the financial information content is analyzed based on the collected user behavior data set, including the following steps: Perform word segmentation and stop word removal preprocessing on the collected financial information text to obtain text data with word form restoration and punctuation removal; Use TF-IDF to identify keywords in financial information that appear more frequently than a preset ratio and are distinguishable in different information texts; The LDA topic model is used to extract topics from the text data of financial information and assign multiple topic tags to each information text.

5. The method for recommending financial information based on multi-channel distribution according to claim 4, characterized in that: When constructing a multi-dimensional user portrait, we analyze the user's interest preferences, active time periods, and reading habits based on the collected user behavior data set, extract the financial information content that the user has read, and annotate it with topic tags in the financial field, including stocks, funds, and bonds.

6. The method for recommending financial information based on multi-channel distribution according to claim 5, characterized in that: Building a multi-dimensional user portrait includes the following steps: Through the collected user behavior data set, statistics are collected on each user's interaction with financial information; Generate an interest model for each user, the interest model calculating the user's preference for a certain type of financial information based on the frequency and time dimension of the information topic tags read by the user; Based on the constructed user profile, collaborative filtering algorithms are used to recommend information that best matches the user's interests; Store each user's profile in the database for subsequent recommendation system calls.

7. The method for recommending financial information based on multi-channel distribution according to claim 6, characterized in that: In S40, when a financial information database is established, the financial information database is a structured database including the following tables: Information table: contains information ID, title, content, release time, category label, and keyword label; Category table: contains category ID, category name, and description; Keyword table: contains keyword ID, keyword name, and related information ID.

8. The method for recommending financial information based on multi-channel distribution according to claim 7, characterized in that: When tagging each piece of financial information with keywords, the RAKE algorithm is used to extract keywords. The specific steps include: segmenting the text and removing stop words, extracting candidate keywords, and obtaining the score of each candidate word by calculating the word frequency, word co-occurrence, and word weight. The keyword with the highest score is selected as the final label. The score is calculated as follows: score = (word frequency × weight) × (co-occurrence frequency).

9. The method for recommending financial information based on multi-channel distribution according to claim 8, characterized in that: S50 is specifically: Collect users' basic information, interest preferences, and historical behavior data, extract users' interest characteristics and behavior characteristics, and form user portraits; Extract keywords and themes from financial information content to form information feature vectors, and classify information content by theme; By calculating the matching degree between the interest keywords in the user portrait and the keywords in the information content, the cosine similarity is used to quantify it, and a matching score is generated for each piece of information; Build a user behavior matrix to record the user's interaction data on information content, use user collaborative filtering to identify users with similar behaviors to the current user, and recommend information that the current user has not been exposed to based on the behaviors of similar users; The keyword matching degree and collaborative filtering recommendation results are weighted and integrated to generate the final recommendation list.

10. A financial information recommendation system based on multi-channel distribution, characterized in that: A system for executing the financial information recommendation method based on multi-channel distribution according to any one of claims 1 to 9, the system comprising: The data collection module is used to collect user behavior data from multiple channels through the front-end system of the financial information platform; Data integration module, used to integrate user behavior data across platforms to form a complete user behavior data set; The portrait generation module is used to analyze the content of financial information based on the collected user behavior data set, extract the topic tags in the financial field, and build a multi-dimensional user portrait; The keyword matching module is used to classify the financial information content on the platform, establish a financial information database, and tag each piece of financial information with a keyword; The information recommendation module is used to recommend financial information that matches the user's preferences by using the matching degree between the user's profile and the keywords in the information content, and to recommend financial information that the user is interested in by using collaborative filtering based on the user's historical behavior data and the behavior habits of user groups with similar interests and behavior patterns, thereby obtaining a list of recommended financial information; The cross-platform distribution module is used to distribute the financial information recommendation list across platforms through multiple channels based on the user's usage habits on different terminals.

Images