Personalized investment recommendation system based on large language model
Analyzing financial news and social media data through large language models to generate market sentiment indexes, and combining traditional factor models to identify new investment factors to generate personalized factor combinations, solving the problem of lack of targeted and adaptable investment portfolios in the existing system, realizing the More accurate and dynamic investment advice.
Patent Information
- Application Number
- CN202411203806.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing investment recommendation system lacks in-depth analysis of users' personalized needs and market sentiment data, resulting in the generated investment portfolios lacking targeted and adaptable, and it is impossible to accurately predict market irrational behavior.
Using a personalized investment recommendation system based on a large language model, analyzing emotional data of financial news, social media information and company announcements generates market sentiment indexes, and combining traditional factor investment models to identify new investment factors and based on user style and risk tolerance. Generate personalized factor combinations, optimize investment strategies, and activate emergency response mechanisms in extreme market events.
It provides more accurate and dynamically adaptable investment advice, which meets user needs and market dynamics, and enhances the accuracy and adaptability of investment decisions.
Smart Images

Figure CN119090534A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of personalized investment recommendation, and in particular, relates to a personalized investment recommendation system based on a large language model. Background Art
[0002] As financial markets become increasingly complex and globalized, investors are facing more and more challenges. Traditional investment advice methods usually rely on historical data and static analysis, which are often unable to cope with the rapid changes in the market. At the same time, factors such as investors' personal needs, risk preferences, and market sentiment fluctuations are not fully considered in traditional methods. In recent years, the development of large language models has brought new opportunities to the financial field. These models have the ability to understand and analyze complex financial information through deep learning of massive text data.
[0003] When generating investment recommendations, existing recommendation systems usually only consider some general investment factors, while ignoring the personalized needs of users, such as investment style, market preference, and risk tolerance. They lack in-depth analysis of these personalized factors and optimization of factor combinations, resulting in the generated investment portfolios often lacking pertinence and personalization, and are unable to fully meet the unique investment goals of users. In addition, most current investment recommendation systems focus on rational analysis based on historical data and structured data, while ignoring market sentiment data from financial news, social media information, and company announcements. Due to the failure to effectively analyze and integrate these sentiment data, existing recommendation systems lack sensitivity and foresight when responding to market sentiment fluctuations, and are unable to accurately predict the irrational behavior of the market, further affecting the accuracy and adaptability of investment decisions.
[0004] In view of this, the present invention is proposed. Summary of the invention
[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a system for personalized investment recommendations based on a large language model, thereby solving the problems raised in the above-mentioned background technology.
[0006] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:
[0007] A personalized investment recommendation system based on a large language model, including: an analysis and evaluation module for sentiment analysis of financial news, social media information and company announcements in market data, generating a market sentiment index by analyzing the sentiment fluctuations of market participants, and incorporating the index into the process of generating investment recommendations; sentiment fluctuations include optimism, pessimism, and neutrality. Data sources include user data, market data, social media, financial news, etc.
[0008] The portfolio module is used to combine the traditional factor investment model and the large language model to optimize the investment strategy by identifying new investment factors in the market and generating personalized factor portfolios based on the user's investment style, market preference and risk tolerance;
[0009] The emergency response module is used to simulate the impact of various extreme market events, generate emergency investment strategies through scenario analysis, and activate the emergency response mechanism when there are sudden market changes, quickly adjust the investment portfolio and issue alerts to users;
[0010] The recommendation result display module is used to display the generated personalized investment advice to users.
[0011] Optionally, the method for constructing and training a large language model module includes the following steps:
[0012] Integrate the data from the data collection module into a data set and divide the data set into a training set and a validation set;
[0013] Use the training set to pre-train the model and design a factor recognition mechanism in the large language model, and use the model's semantic understanding ability to extract new investment factors from market data;
[0014] Integrate sentiment analysis into large language models to analyze sentiment signals in market data. Finally, these signals are converted into market sentiment indices that can be used for investment decisions.
[0015] Use the validation set to validate the model and evaluate its performance in different market scenarios;
[0016] Deploy the trained large language model to the trading platform, enabling it to process user requests and generate investment recommendations in real time.
[0017] Optionally, sentiment analysis of financial news, social media messages, and company announcements in market data can be performed as follows:
[0018] Obtain text data from financial news websites, social media platforms, and company announcements and segment long texts into paragraphs or sentences to form multiple independent text segments;
[0019] Build a sentiment dictionary containing positive, negative, and specific terms related to market sentiment and convert text data into sentiment vectors S i , which is expressed as: Among them, S i represents the sentiment vector of the i-th text segment, w j is the sentiment weight of the jth word in the sentiment dictionary, v j is the word vector of vocabulary j, σ(f(e j )) is the sentiment embedding vector e of word j applied with activation function fj The result after is used to adjust the sentiment intensity of the vocabulary. α is the adjustment factor used to balance the influence between long text and short text. j Represents the frequency of occurrence of word j in the text, which is used to control the contribution of frequently occurring words to the sentiment vector;
[0020] Build a sentiment analysis model in the financial field and train the model using a training set with sentiment labels so that it can identify and classify sentiment tendencies in texts;
[0021] The sentiment of each text segment is classified into positive, negative or neutral categories, and the sentiment score of each text is calculated based on the sentiment classification results. The sentiment scores of all text segments are weighted and summarized to generate the market sentiment index I. t , whose expression is Among them, I t is the market sentiment index at time point t; M is the total number of text fragments at time point t; ω i is the data source weight of text segment i; γ and λ are the sentiment decay coefficient and time decay parameter, respectively, which control the influence of old information on the sentiment index; T i is the release time of text segment i. The sentiment score ranges from -1 to 1, or 0 to 100, where negative sentiment has negative scores, positive sentiment has positive scores, and neutral sentiment is close to zero.
[0022] Optionally, the steps of generating a market sentiment index by analyzing the sentiment fluctuations of market participants and incorporating the index into the generation of investment recommendations are:
[0023] Perform sentiment analysis on financial news, social media information, and company announcements to obtain the sentiment score of each text. i And according to the sentiment score of the obtained text i Calculate the emotional fluctuation ΔS t , whose expression is: Among them, ΔS t represents the emotional fluctuation at time t, N is the number of text segments within time t, and are the sentiment scores of text i in time t and t-1 respectively;
[0024] Based on emotional fluctuations ΔS t , generating the market sentiment index I t , its expression is: t =α·I t-1 +(1-α)·ΔS t , where I t is the market sentiment index at time t; I t-1is the market sentiment index of the previous time period t-1; α is a smoothing parameter that controls the influence weight of current sentiment fluctuations and past indexes.
[0025] Optionally, the steps to optimize the investment strategy are to combine the traditional factor investment model and the large language model by identifying new investment factors in the market and generating a personalized factor portfolio based on the user's investment style, market preference and risk tolerance:
[0026] Select and initialize a set of traditional investment factors to build a factor investment model and use a large language model to conduct in-depth analysis of market data to identify potential new factors that will affect the market;
[0027] Quantify the new factors extracted by the large language model and represent them as factor data that can be used in investment models;
[0028] Screen all candidate factors and use correlation analysis, information ratio and other indicators to evaluate the effectiveness of each factor;
[0029] Based on the user's investment style, market preference and risk tolerance, the weight of each factor is calculated using the following formula: i , which is expressed as: Among them, ω i is the final weight of factor i, γ i is the matching degree between factor i and user’s investment style, InfoRatio i is the information ratio of factor i, which is used to measure the effectiveness of the factor, and M is the total number of effective factors after screening;
[0030] Generate a personalized factor portfolio based on the determined factor weights, form a user-specific investment portfolio strategy, and calculate the expected return and risk of the personalized factor portfolio;
[0031] Apply personalized factor combinations to investment strategy plans and dynamically adjust factor combinations based on real-time market data;
[0032] Generate specific investment recommendations based on personalized factor combinations and optimized investment strategies, and use large language models to provide natural language interpretation of investment recommendations.
[0033] Optionally, the steps to simulate the impact of various extreme market events and generate emergency investment strategies through scenario analysis are as follows:
[0034] Collect samples of extreme market events from historical data and build an extreme event database;
[0035] Use large language models to analyze the characteristics of historical extreme events, generate possible future extreme market event scenarios, and calculate the scenario-adjusted returns and risks of investment portfolios under the scenarios;
[0036] Based on the results of scenario analysis, corresponding emergency investment strategies are generated for different extreme market event scenarios.
[0037] Optionally, the steps to activate the emergency response mechanism in case of sudden market changes, quickly adjust the investment portfolio and issue alerts to users are:
[0038] Monitor market dynamics in real time and detect possible extreme market event signals. Once a sudden market change signal is detected, activate the emergency response mechanism and apply the emergency investment strategy to the current investment portfolio, and quickly adjust asset allocation and risk management measures to respond to sudden market events;
[0039] After executing the emergency response mechanism, the performance of the investment portfolio will be continuously monitored and the strategy will be dynamically adjusted according to further changes in the market.
[0040] Optionally, the steps of presenting the generated personalized investment advice to the user are:
[0041] Generate personalized investment recommendations based on the user's investment style, risk preference, market sentiment index, factor combination and scenario analysis results, and stratify the generated investment recommendations and sort them by importance and priority;
[0042] Leverage large language models to translate technical and data-driven investment recommendations into natural language descriptions that are easy for users to understand;
[0043] After the investment advice is presented to the user, the recommendation system continuously monitors market data and user behavior. If the market changes significantly or the user responds to the advice, the advice content is updated in real time and the presentation method is dynamically adjusted.
[0044] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:
[0045] The present invention optimizes investment strategies and ensures the diversity and personalization of investment portfolios by identifying new investment factors in the market and generating personalized factor combinations based on the user's investment style, market preference and risk tolerance. At the same time, the analysis and evaluation module generates a market sentiment index through sentiment analysis of financial news, social media information and company announcements. The sentiment index adds emotional considerations to investment advice, so that investment decisions are not only based on rational data analysis, but also reflect the emotional fluctuations of the market. Through the synergy of these two modules, the recommendation system can provide more accurate and dynamically adaptive investment advice that is more in line with user needs and the actual dynamics of the market.
[0046] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0048] In the figure:
[0049] Figure 1 System block diagram for personalized investment recommendations.
[0050] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0051] The present invention will now be described in further detail with reference to the accompanying drawings.
[0052] See also Figure 1 As shown, in this embodiment, a system for personalized investment recommendations based on a large language model is provided, including a data collection module for collecting the user's investment history, financial status, risk preference and target return and obtaining external data such as market conditions, financial news, company financial reports, etc. in real time; by collecting the user's investment history, financial status, risk preference and target return, the recommendation system can fully understand the user's investment background and needs.
[0053] The large language model module is used to analyze the processed data and generate personalized investment advice. The large language model module can analyze the processed data and generate highly personalized investment advice through its powerful natural language processing and understanding capabilities.
[0054] The analysis and evaluation module is used to conduct sentiment analysis on financial news, social media information and company announcements in market data. By analyzing the emotional fluctuations of market participants, a market sentiment index is generated and incorporated into the process of generating investment advice. The market sentiment index can reflect the collective sentiment of the market, thereby providing an emotional input for investment strategies and improving the depth and accuracy of investment decisions.
[0055] The combination module is used to combine the traditional factor investment model and the large language model. It optimizes the investment strategy by identifying new investment factors in the market and generating personalized factor combinations based on the user's investment style, market preference and risk tolerance. The factor combination not only covers traditional market indicators, but also captures new market trends and optimizes the performance of investment strategies.
[0056] The emergency response module is used to simulate the impact of various extreme market events, generate emergency investment strategies through scenario analysis, and activate the emergency response mechanism when there are sudden market changes, quickly adjust the investment portfolio and issue an alert to the user; Generate emergency investment strategies through scenario analysis to ensure that when there are sudden changes in the market, the user's investment portfolio can be quickly adjusted to reduce the risk of loss. When the emergency response mechanism identifies sudden market changes, it can immediately activate the emergency response mechanism and issue an alert to the user, allowing the user to take action in the shortest possible time.
[0057] The recommendation result display module is used to display the generated personalized investment advice to users.
[0058] In this embodiment, the method for constructing and training a large language model module includes the following steps:
[0059] Integrate the data from the data collection module into a data set and divide the data set into a training set and a validation set;
[0060] Use the training set to pre-train the model and design a factor recognition mechanism in the large language model, and use the model's semantic understanding ability to extract new investment factors from market data;
[0061] Integrate sentiment analysis into large language models to analyze sentiment signals in market data. Finally, these signals are converted into market sentiment indices that can be used for investment decisions.
[0062] Use the validation set to validate the model and evaluate its performance in different market scenarios;
[0063] The trained large language model is deployed to the trading platform, enabling it to process user requests and generate investment advice in real time. By integrating and allocating different types of data as training sets and validation sets, the model is ensured to be fully trained and tested in a diverse data environment; the integration of factor recognition mechanism and sentiment analysis function enables the model to capture the deep dynamics and sentiment signals of the market, improving the accuracy and real-time nature of investment decisions; finally, the verified model is deployed to the trading platform, ensuring that users can obtain personalized investment advice based on the latest market scenarios in real time.
[0064] In this embodiment, the steps of performing sentiment analysis on financial news, social media information, and company announcements in market data are as follows:
[0065] Obtain text data from financial news websites, social media platforms, and company announcements and segment long texts into paragraphs or sentences to form multiple independent text segments;
[0066] Build a sentiment dictionary containing positive, negative, and specific terms related to market sentiment and convert text data into sentiment vectors Si , which is expressed as: Among them, S i represents the sentiment vector of the i-th text segment, w j is the sentiment weight of the jth word in the sentiment dictionary, v j is the word vector of vocabulary j, σ(f(e j )) is the sentiment embedding vector e of word j applied with activation function f j The result after is used to adjust the sentiment intensity of the vocabulary. α is the adjustment factor used to balance the influence between long text and short text. j Represents the frequency of occurrence of word j in the text, which is used to control the contribution of frequently occurring words to the sentiment vector;
[0067] Build a sentiment analysis model in the financial field and train the model using a training set with sentiment labels so that it can identify and classify sentiment tendencies in texts;
[0068] The sentiment of each text segment is classified into positive, negative or neutral categories, and the sentiment score of each text is calculated based on the sentiment classification results. The sentiment scores of all text segments are weighted and summarized to generate the market sentiment index I. t , whose expression is Among them, I t is the market sentiment index at time point t; M is the total number of text fragments at time point t; ω i is the data source weight of text segment i; γ and λ are the sentiment decay coefficient and time decay parameter, respectively, which control the influence of old information on the sentiment index; T i is the publishing time of text segment i.
[0069] It should be noted that: first, by segmenting paragraphs or sentences, long texts are converted into multiple independent text fragments, ensuring the refined processing of sentiment analysis. Then, a sentiment dictionary containing positive, negative and market sentiment-related terms is constructed, laying a solid foundation for sentiment analysis, and by converting text data into sentiment vectors, the sentiment characteristics of each text fragment can be accurately quantified. By introducing sentiment weights, word vectors, activation functions and adjustment factors, the intensity and tendency of sentiment in the text are accurately captured. At the same time, by controlling the influence of frequently occurring words on the sentiment vector, the balance and stability of sentiment analysis are ensured. Subsequently, the sentiment analysis model unique to the financial field is used, and the model is trained in combination with a training set with sentiment labels, so that it can accurately identify and classify the sentiment tendency in the text, and then sentimentally classify each text fragment, clearly dividing it into positive, negative or neutral categories. Finally, by calculating the sentiment score of each text and weighted summarizing the sentiment scores of all text fragments, a sentiment index reflecting the overall market sentiment is generated. This index not only integrates the weights from different data sources, but also reasonably controls the long-term impact of old information on the sentiment index by introducing sentiment decay coefficient and time decay parameters, thereby achieving dynamic and accurate capture and representation of market sentiment.
[0070] In this embodiment, the steps of generating a market sentiment index by analyzing the emotional fluctuations of market participants and incorporating the index into the process of generating investment advice are as follows:
[0071] Perform sentiment analysis on financial news, social media information, and company announcements to obtain the sentiment score of each text. i And according to the sentiment score of the obtained text i Calculate the emotional fluctuation ΔS t , whose expression is: Among them, ΔS t represents the emotional fluctuation at time t, N is the number of text segments within time t, and are the sentiment scores of text i in time t and t-1 respectively;
[0072] Based on emotional fluctuations ΔS t , generating the market sentiment index I t , its expression is: t =α·I t-1 +(1-α)·ΔS t , where I t is the market sentiment index at time t; I t-1 is the market sentiment index of the previous time period t-1; α is a smoothing parameter that controls the influence weight of current sentiment fluctuations and past indexes.
[0073] It should be noted that by calculating the sentiment score of each text, the changes in market sentiment in different periods are accurately captured. This process ensures real-time monitoring of the emotional fluctuations of market participants. Then, by calculating the emotional fluctuations, the intensity of market sentiment can be quantified, laying the foundation for the subsequent generation of market sentiment index. The generation of the market sentiment index combines the current emotional fluctuations with the historical sentiment index, and uses smoothing parameters to adjust the influence weights of the two, so as to reflect the current market sentiment while retaining the continuity and stability of the historical emotional trend. The setting of the market sentiment index can not only dynamically capture the changes in market sentiment, but also avoid the excessive sensitivity of the market sentiment index due to short-term fluctuations through the smoothing of the index, thereby enhancing the reference value of the index for investment decisions.
[0074] In this embodiment, the traditional factor investment model and the large language model are combined to identify new investment factors in the market and generate personalized factor combinations according to the user's investment style, market preference and risk tolerance. The steps of optimizing the investment strategy are as follows:
[0075] Select and initialize a set of traditional investment factors to establish a factor investment model and use the large language model to conduct in-depth analysis of market data to identify potential new factors that will affect the market; traditional investment factors include price-earnings ratio, momentum factor, and value factor;
[0076] Quantify the new factors extracted by the large language model and express them as factor data that can be used in investment models. Potential new factors include sentiment factors based on market sentiment, macroeconomic factors, and industry-specific factors.
[0077] Screen all candidate factors and use correlation analysis, information ratio and other indicators to evaluate the effectiveness of each factor. The candidate factors include traditional factors and new factors identified by the large language model.
[0078] Based on the user's investment style, market preference and risk tolerance, the weight of each factor is calculated using the following formula: i , which is expressed as: Among them, ω i is the final weight of factor i, γ i is the matching degree between factor i and user’s investment style, InfoRatio i is the information ratio of factor i, which is used to measure the effectiveness of the factor, and M is the total number of effective factors after screening;
[0079] Generate a personalized factor portfolio based on the determined factor weights, form a user-specific investment portfolio strategy, and calculate the expected return and risk of the personalized factor portfolio;
[0080] Apply personalized factor combinations to investment strategy plans and dynamically adjust factor combinations based on real-time market data;
[0081] Generate specific investment recommendations based on personalized factor combinations and optimized investment strategies, and use large language models to provide natural language interpretation of investment recommendations.
[0082] In this embodiment, the steps of simulating the impact of various extreme market events and generating emergency investment strategies through scenario analysis are as follows:
[0083] Collect samples of extreme market events from historical data and build an extreme event database;
[0084] Use large language models to analyze the characteristics of historical extreme events, generate possible future extreme market event scenarios, and calculate the scenario-adjusted returns and risks of investment portfolios under the scenarios;
[0085] According to the results of scenario analysis, corresponding emergency investment strategies are generated for different extreme market event scenarios. It should be noted that the foundation of the factor investment model is laid by selecting and initializing a set of traditional investment factors. At the same time, the large language model is used to conduct in-depth analysis of market data to identify emerging factors that may affect the market, which enables the model to capture new trends and potential opportunities in the market. Subsequently, these new factors are quantified and converted into actionable factor data to ensure that the model can make full use of these innovative factors. All candidate factors are strictly screened through indicators such as correlation analysis and information ratio to ensure the effectiveness and relevance of the factors. Next, based on the user's investment style, market preference and risk tolerance, the weight of each factor is calculated to ensure that the final generated factor portfolio is highly matched with the user's needs. In particular, the calculation formula of the factor weight not only takes into account the effectiveness of the factor, but also combines the user's investment style to ensure the personalization and accuracy of the factor portfolio. According to the determined factor weights, a personalized factor portfolio exclusive to the user is generated, forming an investment portfolio strategy that is highly matched with the user's specific needs. At the same time, by calculating the expected return and risk, a comprehensive investment performance expectation is provided to the user. In actual applications, the recommendation system can also combine real-time market data to dynamically adjust factor combinations to ensure the flexibility and timeliness of investment strategies. Finally, based on personalized factor combinations and optimized investment strategies, the recommendation system generates specific investment recommendations and uses large language models to convert these recommendations into natural language, allowing users to easily understand and adopt these recommendations.
[0086] In this embodiment, the steps of activating the emergency response mechanism when a sudden market change occurs, quickly adjusting the investment portfolio and issuing an alert to the user are as follows:
[0087] Monitor market dynamics in real time and detect possible extreme market event signals. Once a sudden market change signal is detected, activate the emergency response mechanism and apply the emergency investment strategy to the current investment portfolio, and quickly adjust asset allocation and risk management measures to respond to sudden market events;
[0088] After the emergency response mechanism is executed, the performance of the portfolio is continuously monitored and the strategy is dynamically adjusted according to further changes in the market. By monitoring market dynamics in real time and detecting sudden signals, the emergency response mechanism can be quickly activated, so that asset allocation and risk management measures can be adjusted in a timely manner. The mechanism of immediate response ensures that when the market fluctuates violently, the portfolio can quickly adapt to new market conditions and minimize potential risks. By quickly applying emergency investment strategies to the current portfolio, the recommendation system can effectively protect the safety of investors' assets. In addition, after the emergency response mechanism is executed, the recommendation system will continue to monitor the performance of the portfolio and dynamically adjust the strategy according to further changes in the market. The ability to continuously monitor and dynamically adjust ensures the flexibility and adaptability of the investment strategy in extreme situations, thereby improving the robustness and risk resistance of the portfolio.
[0089] In this embodiment, the steps of presenting the generated personalized investment advice to the user are:
[0090] Generate personalized investment recommendations based on the user's investment style, risk preference, market sentiment index, factor combination and scenario analysis results, and stratify the generated investment recommendations and sort them by importance and priority;
[0091] Leverage large language models to translate technical and data-driven investment recommendations into natural language descriptions that are easy for users to understand;
[0092] After the investment advice is presented to the user, the recommendation system continuously monitors market data and user behavior. If the market changes significantly or the user responds to the advice, the advice content is updated in real time and the display method is adjusted dynamically. The present invention uses a large language model to convert these complex technical and data-driven advice into natural language descriptions that are easy for users to understand, thereby enhancing the user's understanding and decision-making ability. In addition, the recommendation system continues to monitor market dynamics and user behavior after the advice is presented, ensuring that the advice content is updated in real time and the display method is adjusted dynamically when the market changes significantly or the user responds. The mechanism of dynamic adjustment and real-time response not only improves the accuracy and timeliness of investment advice, but also enhances the user experience, making investment decisions more flexible, reliable and in line with personalized needs.
[0093] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.
Claims
1. A personalized investment recommendation system based on a large language model, characterized in that: include: Data collection module, which is used to collect users' investment history, financial status, risk preferences and target returns, and obtain external data such as market conditions, financial news and company financial reports in real time; A large language model module, used to analyze the processed data and generate personalized investment advice; The analysis and evaluation module is used to perform sentiment analysis on financial news, social media information and company announcements in market data, generate a market sentiment index by analyzing the sentiment fluctuations of market participants, and incorporate the index into the process of generating investment recommendations; The portfolio module is used to combine the traditional factor investment model and the large language model to optimize the investment strategy by identifying new investment factors in the market and generating personalized factor portfolios based on the user's investment style, market preference and risk tolerance; The emergency response module is used to simulate the impact of various extreme market events, generate emergency investment strategies through scenario analysis, and activate the emergency response mechanism when there are sudden market changes, quickly adjust the investment portfolio and issue alerts to users; The recommendation result display module is used to display the generated personalized investment advice to users.
2. A system for personalized investment recommendation based on a large language model according to claim 1, characterized in that: The construction and training method of the large language model module includes the following steps: Integrate the data from the data collection module into a data set and divide the data set into a training set and a validation set; Use the training set to pre-train the model and design a factor recognition mechanism in the large language model, and use the model's semantic understanding ability to extract new investment factors from market data; Integrate sentiment analysis into large language models to analyze sentiment signals in market data. Finally, these signals are converted into market sentiment indices that can be used for investment decisions. Use the validation set to validate the model and evaluate its performance in different market scenarios; Deploy the trained large language model to the trading platform, enabling it to process user requests and generate investment recommendations in real time.
3. A system for personalized investment recommendation based on a large language model according to claim 1, characterized in that: The steps for performing sentiment analysis on financial news, social media information, and company announcements in market data are: Obtain text data from financial news websites, social media platforms, and company announcements and segment long texts into paragraphs or sentences to form multiple independent text segments; Build a sentiment dictionary containing positive, negative, and specific terms related to market sentiment and convert text data into sentiment vectors S i , which is expressed as: Among them, S i represents the sentiment vector of the i-th text segment, w j is the sentiment weight of the jth word in the sentiment dictionary, v j is the word vector of vocabulary j, σ(f(e j )) is the sentiment embedding vector e of word j applied with activation function f j The result after is used to adjust the sentiment intensity of the vocabulary. α is the adjustment factor used to balance the influence between long text and short text. j Represents the frequency of occurrence of word j in the text, which is used to control the contribution of frequently occurring words to the sentiment vector; Build a sentiment analysis model in the financial field and train the model using a training set with sentiment labels so that it can identify and classify sentiment tendencies in texts; The sentiment of each text segment is classified into positive, negative or neutral categories, and the sentiment score of each text is calculated based on the sentiment classification results. The sentiment scores of all text segments are weighted and summarized to generate the market sentiment index I. t , whose expression is Among them, I t is the market sentiment index at time point t; M is the total number of text fragments at time point t; ω i is the data source weight of text segment i; γ and λ are the sentiment decay coefficient and time decay parameter, respectively, which control the influence of old information on the sentiment index; T i is the publishing time of text segment i.
4. A system for personalized investment recommendation based on a large language model according to claim 1, characterized in that: The steps to generate a market sentiment index by analyzing the emotional fluctuations of market participants and incorporating the index into the generation of investment recommendations are: Perform sentiment analysis on financial news, social media information, and company announcements to obtain the sentiment score of each text. i And according to the sentiment score of the obtained text i Calculate emotional fluctuations △S t , whose expression is: Among them, △S t represents the emotional fluctuation at time t, N is the number of text segments within time t, and are the sentiment scores of text i in time t and t-1 respectively; Based on emotional fluctuations △S t , generating the market sentiment index I t , its expression is: t =α·I t-1 +(1-α)·△S t , where I t is the market sentiment index at time t; I t-1 is the market sentiment index of the previous time period t-1; α is a smoothing parameter that controls the influence weight of current sentiment fluctuations and past indexes.
5. The system for personalized investment recommendation based on a large language model according to claim 1, characterized in that: Combining the traditional factor investment model and the big language model, by identifying new investment factors in the market and generating personalized factor combinations based on the user's investment style, market preference and risk tolerance, the steps to optimize the investment strategy are: Select and initialize a set of traditional investment factors to build a factor investment model and use a large language model to conduct in-depth analysis of market data to identify potential new factors that will affect the market; Quantify the new factors extracted by the large language model and represent them as factor data that can be used in investment models; Screen all candidate factors and use correlation analysis and information ratio indicators to evaluate the effectiveness of each factor; Based on the user's investment style, market preference and risk tolerance, the weight of each factor is calculated using the following formula: i , which is expressed as: Among them, ω i is the final weight of factor i, γ i is the matching degree between factor i and user’s investment style, InfoRatio i is the information ratio of factor i, which is used to measure the effectiveness of the factor, and M is the total number of effective factors after screening; Generate a personalized factor portfolio based on the determined factor weights, form a user-specific investment portfolio strategy, and calculate the expected return and risk of the personalized factor portfolio; Apply personalized factor combinations to investment strategy plans and dynamically adjust factor combinations based on real-time market data; Generate specific investment recommendations based on personalized factor combinations and optimized investment strategies, and use large language models to provide natural language interpretation of investment recommendations.
6. A system for personalized investment recommendation based on a large language model according to claim 1, characterized in that: The steps to simulate the impact of various extreme market events and generate emergency investment strategies through scenario analysis are as follows: Collect samples of extreme market events from historical data and build an extreme event database; Use large language models to analyze the characteristics of historical extreme events, generate future extreme market event scenarios, and calculate the scenario-adjusted returns and risks of investment portfolios under the scenarios; Based on the results of scenario analysis, corresponding emergency investment strategies are generated for different extreme market event scenarios.
7. A system for personalized investment recommendation based on a large language model according to claim 1, characterized in that: The steps to activate the emergency response mechanism when sudden market changes occur, quickly adjust the investment portfolio and issue alerts to users are: Monitor market dynamics in real time and detect signals of extreme market events. Once a sudden market change signal is detected, activate the emergency response mechanism and apply the emergency investment strategy to the current investment portfolio, and quickly adjust asset allocation and risk management measures to respond to sudden market events; After executing the emergency response mechanism, the performance of the investment portfolio will be continuously monitored and the strategy will be dynamically adjusted according to further changes in the market.
8. The system for personalized investment recommendation based on a large language model according to claim 1, characterized in that: The steps to present the generated personalized investment advice to the user are: Generate personalized investment recommendations based on the user's investment style, risk preference, market sentiment index, factor combination and scenario analysis results, and stratify the generated investment recommendations and sort them by importance and priority; Leverage large language models to translate technical and data-driven investment recommendations into natural language descriptions that are easy for users to understand; After the investment advice is presented to the user, the recommendation system continuously monitors market data and user behavior. If the market changes significantly or the user responds to the advice, the advice content is updated in real time and the presentation method is dynamically adjusted.
Citation Information
Cited By
Financial user emotion and psychological portrait analysis method based on large language model
CN119293251A
Intelligent pushing method of security information
CN119477547A
