Device for communicating with user based on credit card transaction records to recommend products and method thereof

TWI945652BActive Publication Date: 2026-11-11TAIWAN BUSINESS BANK
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
TW113151570
Authority / Receiving Office
TW · TW
Patent Type
Patents
Current Assignee / Owner
Filing Date
2024-12-30
Publication Date
2026-11-11
Estimated Expiration
2044-12-29

AI Technical Summary

Technical Problem

Current banking systems fail to accurately recommend products to customers due to a lack of interaction, resulting in low customer willingness to purchase recommended products.

Method used

A system and method that analyzes historical credit card swipe records and user dialogue using a generative language model to adjust product recommendations based on user interactions, incorporating a data acquisition, trend analysis, demand analysis, weight adjustment, and real-time interaction modules.

Benefits of technology

Enhances the relevance of product recommendations by understanding customer needs, thereby increasing the willingness of bank customers to purchase recommended products.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure TWG2TB001920397_001
    Figure TWG2TB001920397_001
  • Figure TWG2TB001920397_002
    Figure TWG2TB001920397_002
  • Figure TWG2TB001920397_003
    Figure TWG2TB001920397_003
Patent Text Reader

Abstract

A system and method for recommending products based on card swipe records and user dialogue involves analyzing historical card swipe data of bank customers through a bank server to generate consumption trend information. Then, a generative language model is used to generate dialogue messages, enabling the bank application to interact with the user via voice or text. The system adjusts the weight values ​​of mainstream products in the consumption trend information based on the user interaction content, and determines the mainstream products to recommend to the user based on the adjusted weight values. This allows the bank to recommend products that meet the purchasing needs of its customers, thereby increasing the customers' willingness to purchase the products recommended by the bank.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] A product recommendation system and method thereof, particularly a system and method for recommending products based on card swipe records and conversations with users. Prior Technology

[0002] Currently, the banking industry typically uses customer preferences and / or historical transaction records, among other data, to generate financial products recommended to customers based on predetermined rules.

[0003] However, the aforementioned product recommendation method is one-sided. The bank does not communicate with its customers, resulting in the bank not truly understanding the customers' purchasing needs. Most customers can only passively choose to accept or reject the recommended products, which leads to a generally low willingness among bank customers to purchase the products recommended by the bank.

[0004] In summary, it is clear that previous technologies have long suffered from the problem that the products recommended by banks to their customers do not meet the customers' purchasing needs. Therefore, it is necessary to propose improved technical means to solve this problem. Summary of the Invention

[0005] In view of the problem that prior art has the issue that the products recommended by banks to their customers do not meet the customers' purchasing needs, the present invention discloses a system and method for recommending products based on card swipe records and user dialogue, wherein:

[0006] The system disclosed in this invention, which recommends products based on credit card swipe records and user conversations, includes at least: a data acquisition module for acquiring historical credit card swipe records of multiple bank customers; a trend analysis module for using a computational model to analyze the historical credit card swipe data of bank customers to generate consumption trend information, which includes product information of multiple mainstream products; a demand analysis module for analyzing the user's bank customer data to generate potential demand, where the user is one of the bank customers and the potential demand includes desired products; a weight adjustment module for setting weight values ​​for mainstream products based on the similarity of desired products to mainstream products; and a model maintenance module for providing product information and weight values ​​of mainstream products to the generative algorithm. The language model consists of a generative language model trained on pre-established interactive examples; a real-time interaction module that uses the generative language model to generate multiple dialogue messages and send them to the client, allowing the client to continuously interact with the user by outputting dialogue messages via voice and / or text; a weight adjustment module that adjusts the weight values ​​of mainstream products based on user responses during the interaction, and a model maintenance module that provides the adjusted weight values ​​of mainstream products to the generative language model, enabling the generative language model to generate dialogue messages based on these weight values; and a product recommendation module that determines recommended products based on the weight values ​​of mainstream products, allowing the client to recommend products to the user.

[0007] The method for recommending products based on credit card swipe records and user dialogue disclosed in this invention includes at least the following steps: a bank server obtains historical credit card swipe records of multiple bank customers; the bank server uses a computational model to analyze the historical credit card swipe data of bank customers to generate consumption trend information, which includes product information of multiple mainstream products; the bank server analyzes the user's bank customer data to generate potential needs, where the user is one of the bank customers and the potential needs include desired products; the bank server assigns weight values ​​to mainstream products based on the similarity between desired products and mainstream products; the bank server provides product information and weight values ​​of mainstream products to a generative language model, which then... The language model is trained using pre-established interactive examples; the bank server uses the generative language model to generate multiple dialogue messages and sends them to the bank application, enabling the bank application to continuously interact with the user by outputting dialogue messages in voice and / or text. During user interaction, the bank server adjusts the weights of mainstream products based on the user's responses and provides these adjusted weights to the generative language model, which then generates dialogue messages based on these weights. The bank server determines recommended products based on the weights of the mainstream products, and the bank application recommends these products to the user.

[0008] The system and method disclosed in this invention are as described above. The difference between this invention and the prior art is that this invention analyzes the historical card-swiping data of bank customers through a bank server to generate consumption trend information. Then, it uses a generative language model to generate dialogue messages so that the bank application can use voice or text to output dialogue messages to interact with the user. Based on the content of the interaction with the user, the weight values ​​of mainstream products in the consumption trend information are adjusted, and based on the adjusted weight values ​​of mainstream products, mainstream products recommended for the user to purchase are determined. This solves the problems existing in the prior art and can achieve the technical effect of increasing the willingness of bank customers to purchase products recommended by the bank. Simple Explanation of the Diagram

[0009] Figure 1 is a schematic diagram of the components of the device proposed in this invention that recommends products based on card swipe records and conversations with the user. Figure 2 is a schematic diagram of the processor module proposed in this invention. Figure 3A is a flowchart of the method proposed in this invention for recommending products based on card swipe records and user conversations. Figure 3B is a flowchart of the method for generating consumption trend information based on historical credit card swipe records proposed in this invention. Figure 3C is a flowchart of the method for analyzing external bank data to add consumer trend information proposed in this invention. Figure 3D is a flowchart of the method for providing recommended products for purchase according to the present invention. Implementation

[0010] The features and implementation methods of the present invention will be described in detail below with reference to the accompanying drawings and embodiments. The content is sufficient to enable anyone skilled in the art to easily and fully understand the technical means used by the present invention to solve the technical problem and to implement it accordingly, thereby achieving the effects that the present invention can achieve.

[0011] This invention allows a bank server to identify mainstream products based on credit card spending and can use a generative language model to generate dialogue messages to interact with users, thereby selecting recommended products to users based on their responses during the interaction.

[0012] The bank server implementing this invention can be a computing device. This computing device includes, but is not limited to, one or more processing modules, one or more memory modules, and buses connecting different hardware components (including memory modules and processing modules). Through these hardware components, the computing device can load and execute an operating system, allowing the operating system to run on the computing device, and can also execute software or programs. The computing device also includes a housing, within which the aforementioned hardware components are housed.

[0013] The bus of the computing device proposed in this invention may include one or more types, such as data bus, address bus, control bus, expansion bus, and / or local bus. The bus of the computing device includes, but is not limited to, Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, Video Electronics Standards Association (VESA) local buses, and serial Universal Serial Bus (USB), PCI Express (PCI-E / PCIe) buses, etc.

[0014] The processing module of the computing device proposed in this invention is coupled to a bus. The processing module includes a register set or register space, which may be entirely located on the processing chip of the processing module, or wholly or partially located outside the processing chip and coupled to the processing chip via dedicated electrical connections and / or via a bus. The processing module may be a central processing unit, a microprocessor, or any suitable processing element. If the computing device is a multiprocessor device, that is, the computing device contains multiple processing modules, then the processing modules contained in the computing device are identical or similar and are coupled and communicate via a bus. In some embodiments, the processing module can interpret a computer instruction or a series of multiple computer instructions to perform specific calculations or operations, such as mathematical operations, logical operations, data comparison, copying / moving data, etc., thereby driving other hardware components in the computing device or running an operating system or executing various programs and / or modules. Computer instructions can be assembly language instructions, instruction set architecture instructions, machine instructions, machine-dependent instructions, microinstructions, firmware instructions, or source code or object code written in any combination of one or more programming languages. Computer instructions can be executed entirely on a single computing device, partially on a single computing device, or partially on one computing device and partially on another connected computing device. The aforementioned programming languages ​​include object-oriented programming languages ​​such as Common Lisp, Python, C++, Objective-C, Smalltalk, Delphi, Java, Swift, C#, Perl, Ruby, etc., and conventional procedural programming languages ​​such as C or other similar programming languages.

[0015] Computing devices typically include one or more chipsets. The processing modules of the computing device can be coupled to or electrically connected to the chipset via buses. A chipset consists of one or more integrated circuits (ICs), including memory controllers and peripheral input / output (I / O) controllers. That is, the memory controller and I / O controller can be contained within a single IC or implemented using two or more ICs. Chipsets typically provide I / O and memory management functions, as well as multiple general-purpose and / or dedicated registers, timers, etc., which can be accessed or used by one or more processing modules coupled or electrically connected to the chipset. In some embodiments, the chipset may also be part of the processing module.

[0016] The processing module of a computing device can also access data in the memory modules and mass storage areas installed on the computing device through the memory controller. The aforementioned memory modules include any type of volatile memory and / or non-volatile memory (NVRAM), such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Read-Only Memory (ROM), and Flash memory. The aforementioned mass storage areas can contain any type of storage device or storage media, such as hard drives, optical discs, flash drives, memory cards, solid-state drives (SSDs), or any other storage devices. In other words, the memory controller can access data in static random access memory, dynamic random access memory, flash memory, hard drives, and solid-state drives.

[0017] The processing module of a computing device can also connect and communicate with peripheral devices or interfaces such as peripheral output devices, peripheral input devices, communication interfaces, and various data or signal receiving devices via a peripheral input / output controller and a peripheral input / output bus. Peripheral input devices can be any type of input device, such as a keyboard, mouse, trackball, touchpad, joystick, etc., and peripheral output devices can be any type of output device, such as a monitor, printer, etc. Peripheral input devices and peripheral output devices can also be the same device, such as a touch screen. The communication interface can include wireless and / or wired communication interfaces. Wireless communication interfaces can include interfaces supporting wireless local area networks (such as Wi-Fi, Zigbee, etc.), Bluetooth, infrared, near-field communication (NFC), 3G / 4G / 5G mobile communication networks (cellular networks), or other wireless data transmission protocols. Wired communication interfaces can be Ethernet devices, DSL modems, cable modems, asynchronous transfer mode (ATM) devices, or fiber optic communication interfaces and / or components. Data or signal receiving devices can include GPS receivers or physiological signal receivers. Physiological signals received by the physiological signal receiver include, but are not limited to, heart rate and blood oxygenation. The processing module can periodically poll various peripheral devices and interfaces, enabling the computing device to input and output data through these peripheral devices and interfaces, and also to communicate with another computing device that has the hardware components described above.

[0018] The following description uses the component diagram of Figure 1, which illustrates the device for recommending products based on card swipe records and user interaction, to illustrate the implementation of the present invention. As shown in Figure 1, the device 100 of the present invention includes a memory 110, a communication interface 130, a storage medium 140, a processor 170, and a bus 190. The processor 170 is connected to the memory 110, the communication interface 130, and the storage medium 140 via the bus 190.

[0019] Memory 110 can store one or more sets of computer instructions.

[0020] The communication interface 130 can connect to external network storage devices or servers (not shown in the figure) and request and download data from the connected network devices.

[0021] Storage medium 140 can store data or signals downloaded by storage communication interface 130, data or signals provided to processor 170 or required for processor 170 to operate, and data or signals generated by processor 170.

[0022] The processor 170, as shown in the schematic diagram of the modules proposed in Figure 2, includes modules such as a data acquisition module 210, a trend analysis module 220, a demand analysis module 230, a weight adjustment module 240, a model maintenance module 250, a real-time interaction module 260, and a product recommendation module 270. In some embodiments, the processor 170 can execute computer instructions stored in the memory 110 and generate the modules shown in Figure 2 after executing the computer instructions; in other embodiments, the modules shown in Figure 2 can be generated by one or more circuits and / or complete or partial chips and other hardware components, that is, the processor 170 includes the hardware components that make up the modules in Figure 2. In other words, the modules included in the processor 170 can be software modules or hardware modules, and the present invention does not have any particular limitations.

[0023] Data acquisition module 210 is responsible for acquiring the historical card transaction records of bank customers. The historical card transaction records acquired by data acquisition module 210 are multiple card transaction records of all bank customers or multiple specific bank customers within a certain time range. Each card transaction data includes, but is not limited to, the type of credit card used, the time of the transaction, the location of the transaction, the amount of the transaction, the type of goods purchased, and the category of the merchant.

[0024] The data acquisition module 210 can connect to the bank's internal credit card transaction server or other data storage server through the communication interface 130 to download historical card swipe records, and can also read pre-stored historical card swipe records from the storage medium 140. There are no particular limitations to this invention.

[0025] The data acquisition module 210 can also obtain information of interest through social media platforms, third-party analysis reports, and / or search engines. For example, the data acquisition module 210 can use web crawlers or application programming interfaces (APIs) provided by social media platforms to collect content publicly discussed on one or more social media platforms and use the collected discussion content as information of interest; the data acquisition module 210 can also use web crawlers to collect reviews of various products on one or more online sales channels (such as various online shopping platforms) and use the collected reviews as information of interest; the data acquisition module 210 can also download research reports, market analysis / survey reports, economic analysis reports, stock market / interest rate / foreign exchange analysis reports, competitive product analysis reports, and other analysis reports published by various institutions or organizations and use the downloaded analysis reports as information of interest; the data acquisition module 210 can also obtain keywords with a certain level of popularity through application programming interfaces provided by search engines and use the obtained keywords as information of interest. However, the information of interest and the methods of obtaining them mentioned in this invention are not limited to the above.

[0026] Trend analysis module 220 is responsible for analyzing the historical card-swiping data obtained by data acquisition module 201 to generate consumption trend information. The consumption trend information generated by trend analysis module 220 includes product information for one or more mainstream products.

[0027] Generally, the trend analysis module 220 can use one or more computational models to analyze historical credit card transaction data. For example, the trend analysis module 220 can use clustering analysis models such as k-means or DBSCAN to classify various credit card transactions in historical transaction records based on one or more of the following: transaction merchant, consumption behavior, investment preferences, transaction time, and transaction location. The trend analysis module 220 can also use time series analysis models such as ARIMA or Prophet to analyze whether there are transaction trends (such as seasonal or regional consumption) for various credit card transactions based on the transaction time and / or transaction location in historical transaction records. The trend analysis module 220 can also use association analysis models such as Apriori or FP-Growth to find the transaction correlation between combinations of goods and / or consumption behaviors that appear a certain number of times in various credit card transactions and various credit card transactions. The trend analysis module 220 can calculate the index scores of various credit card payment products based on the weights defined in the above-mentioned cluster analysis model, time series analysis model, correlation analysis model, and other calculation models. It can also sum up the index scores of various credit card payment products in terms of classification, transaction trend, product combination, and transaction correlation to generate a total index score. It can sort various credit card payment products according to the calculated total index score, select the products with the highest ranking as mainstream products, and obtain the product information of mainstream products through the communication interface 130 to generate consumption trend information.

[0028] The trend analysis module 220 can use a Natural Language Processing (NLP) model and the attention information obtained by the data acquisition module to extract popular products, and add the extracted popular product information to the consumer trend information. For example, the trend analysis module 220 can use word segmentation techniques such as TF-IDF, word frequency analysis, and TextRank to decompose the attention information such as content and research results collected by the data acquisition module 210 into multiple words and count the number of each word to obtain popular words. It can then use techniques such as Latent Dirichlet Allocation (LDA) based on these popular words. LDA, BERT, and other topic-building technologies segment the information into paragraphs with different themes. They can also re-count the number of words in each paragraph within different themes and determine whether a word whose count reaches a threshold is the name of a product. If so, the trend analysis module 220 can treat that word as a popular product and obtain product information for that product through the communication interface 130. This product information is then added to the generated consumer trend information. The trend analysis module 220 can also use natural language processing technologies such as sentiment analysis to analyze statistical data collected by module 210. The review messages (attention messages) of each product can be positive, neutral, or negative. When the number of positive review messages for a product reaches a positive threshold, the product can be designated as a popular product and its information can be added to the consumer trend information. The trend analysis module 220 can also determine whether the popular keywords (attention messages) obtained by the data acquisition module 210 are the product names of a specific product. When the popular keywords are product names, the module can obtain the product information of the product represented by the keywords through the communication interface 130 and add the obtained product information to the consumer trend information.

[0029] In some embodiments, when analyzing historical card-swipe data, the trend analysis module 220 can also delete card-swipe records with excessively high or low transaction amounts. For example, it can delete card-swipe records with transaction amounts exceeding three standard deviations. It can also delete or retain specific card-swipe records based on customer spending habits. For example, it can retain card-swipe records with similar spending times, similar spending locations, and the same spending type that reach a certain threshold. However, the present invention is not limited to the above.

[0030] The demand analysis module 230 is responsible for analyzing customer data of bank customers (also referred to as bank customer data in this invention) to generate potential demands. The potential demands generated by the demand analysis module 230 include one or more desired products. Generally, the demand analysis module 230 analyzes one bank customer at a time. The bank customer data analyzed by the demand analysis module 230 includes, but is not limited to, the bank customer's interests, deposit amount, credit card transaction history, investment history, and credit limit. Generally, the demand analysis module 230 can connect to one or more service hosts within the bank via the communication interface 130 to obtain bank customer data.

[0031] The demand analysis module 230 can use collaborative filtering algorithms to analyze bank customer data to generate potential demand. For example, the demand analysis module 230 can use user-based collaborative filtering, which means it can find one or more other bank customer data that have a customer similarity value that reaches a customer threshold value with the bank customer data being analyzed, and extract the card-swiping records of the found other bank customer data, and use the extracted card-swiping records as potential demand; the demand analysis module 230 can also use item-based collaborative filtering, which means it can extract card-swiping records from the bank customer data being analyzed, and obtain one or more other products that have a product similarity value that reaches a product threshold value with the extracted card-swiping records, and use the obtained products as potential demand.

[0032] The weight adjustment module 240 is responsible for judging the similarity between each mainstream product generated by the trend analysis module 220 and each demand product generated by the demand analysis module 230, and setting the weight value of each mainstream product based on the sum of the similarity between each mainstream product and each demand product. The similarity of the same mainstream product is directly proportional to the weight value. That is, the higher (or lower) the similarity between the mainstream product and the demand product, the higher (or lower) the weight value of the mainstream product.

[0033] The weight adjustment module 240 is also responsible for adjusting the weight values ​​of mainstream products in the generative language model based on one or more response messages obtained by the real-time interaction module 260 during user interactions with the real-time interaction module 260. For example, when the response message is a positive response or inquiry about a mainstream product, or mentions a mainstream product in the response message, the weight adjustment module 240 can provide the generative language model with instructions or statements to increase the weight value of the mainstream product that received a positive response, inquiry, or mention, thereby allowing the generative language model to increase the weight value of the mainstream product that received a positive response, inquiry, or mention. The weight adjustment module 240 can also provide the generative language model with instructions or statements to delete the mainstream product that received a negative response when the response message is a negative response to a mainstream product, thereby allowing the generative language model to remove the mainstream product that received a negative response from the set mainstream products, or provide the generative language model with instructions or statements to reduce the weight value of the mainstream product that received a negative response to the lowest possible level, thereby allowing the generative language model to increase the weight value of the mainstream product that received a positive response, or reduce the weight value of the mainstream product that received a negative response to the lowest possible level. The model reduces the weight of mainstream products that receive negative responses to zero or negative values. The weight adjustment module 240 can also provide instructions or statements to the generative language model to reduce the weight of mainstream products that did not receive responses when the response message does not address a particular mainstream product. Furthermore, when the response message is a rating or opinion for a mainstream product, the weight adjustment module 240 can use a natural language processing model to analyze the rating or opinion to generate analysis results. Based on the semantics of the analysis results, it can select between positive and negative biases to provide instructions or statements to the generative language model to increase or decrease the weight of mainstream products that received ratings or opinions. However, the methods by which the weight adjustment module 240 adjusts the weight of mainstream products are not limited to the above.

[0034] The model maintenance module 250 can train a generative language model using pre-established interactive paradigms. The generative language model proposed in this invention includes, but is not limited to, GPT-4.

[0035] The model maintenance module 250 is also responsible for providing the mainstream products in the consumption trend information generated by the trend analysis module 220 and the weight values ​​of the mainstream products generated or adjusted by the weight adjustment module 240 to the generative language model, so that the generative language model can generate dialogue messages that recommend mainstream products based on the weight values ​​of mainstream products. In other words, the generative language model can determine the priority recommendation and main recommendation of mainstream products based on the weight values ​​of mainstream products and generate corresponding dialogue messages.

[0036] The real-time interaction module 260 is responsible for generating multiple dialogue messages using the generative language model maintained by the model maintenance module 250, and transmitting the generated dialogue messages to the client 101 through the communication interface 130, so that the client 101 can output dialogue messages in voice and / or text to continuously interact with the user. For example, the real-time interaction module 260 can use the generative language model to generate an opening dialogue message and transmit the opening dialogue message to the client 101 through the communication interface 130. The real-time interaction module 260 can also receive the response message from the client 101 corresponding to the opening dialogue message through the communication interface 130 and provide the response message to the generative language model, so that the generative language model generates another dialogue message based on the response message. The real-time interaction module 260 can then transmit the newly generated dialogue message to the client 101 again through the communication interface 130, and so on.

[0037] The real-time interaction module 260 is also responsible for providing the response messages obtained during the interaction with the user to the weight adjustment module 240. That is, it provides the response messages received from the client 101 through the communication interface 130 to the weight adjustment module 240, so that the weight adjustment module 240 adjusts the weight values ​​of the mainstream products according to the response messages, and then the generative language model generates dialogue messages to be sent to the client 101 based on the adjusted weight values ​​of the mainstream products.

[0038] The product recommendation module 270 is responsible for determining recommended products based on the weight values ​​of mainstream products in the generative language model maintained by the model maintenance module 250. It can also transmit the product information of the recommended products to the client 101 via the communication interface 130, allowing the client 101 to recommend the products to the user. The product recommendation module 270 can determine a mainstream product as a recommended product when its weight value reaches a recommendation threshold, or it can determine the mainstream product with the highest weight value as the recommended product after interacting with the user for a certain period. However, the method by which the product recommendation module 270 determines recommended products based on the weight values ​​of mainstream products is not limited to the above.

[0039] In some embodiments, the product recommendation module 270 can also provide the client 101 with the option to purchase recommended mainstream products through designated sales channels.

[0040] The operating system and method of the present invention will be explained next with an embodiment, and please refer to Figure 3A, which is a flowchart of the method for recommending products based on card swipe records and user dialogue. In the present invention, it is assumed that device 100 is a bank server, and client 101 can be a bank application or a computing device that runs a bank application.

[0041] First, the data acquisition module 210 of device 100 can obtain the historical card swipe records of multiple bank customers (step 301). In this embodiment, it is assumed that the data acquisition module 210 can connect to the bank's internal credit card transaction server through the communication interface 130 to request and download the historical card swipe records of all bank customers in the past three years.

[0042] After the data acquisition module of device 100 acquires the historical card swiping records of bank customers (step 301), the trend analysis module 220 of device 100 can use one or more calculation models to analyze the historical card swiping data of bank customers to generate consumption trend information (step 310). In this embodiment, assuming that the trend analysis module 220 is as shown in the flowchart of "Figure 3B", it can use a cluster analysis model to classify each card swipe item in the historical card swipe records according to the transaction merchant, consumption behavior, investment preference, transaction time, or transaction location (step 311), and can use a time series analysis model to analyze the transaction trend of each card swipe item according to the transaction time and / or transaction location in the historical card swipe records (step 313). It can also use a correlation analysis model to find combinations of goods that appear a certain number of times in each card swipe item and / or the transaction correlation between consumption behavior and card swipe items in the historical card swipe records (step 313). Step 315), then the trend analysis module 220 can also calculate the index scores of each credit card transaction item for category, transaction trend, product combination, and transaction correlation, and adjust the index scores of each credit card transaction item for category, transaction trend, product combination, and transaction correlation according to the weights of the cluster analysis model, time series analysis model, and correlation analysis model, and then sum up the index scores to generate a total index score (step 317). Then, the credit card transaction items are sorted according to the total index score, and a certain number (such as 5 or 10) of the top-ranked credit card transaction items are selected as mainstream items to generate consumption trend information (step 319).

[0043] Returning to Figure 3A, after the trend analysis module 220 of device 100 generates consumption trend information, the demand analysis module 230 of device 100 can analyze the bank customer data of the user of client 101 to generate the user's potential needs (step 330). In this embodiment, assuming the user is one of the bank customers, the demand analysis module 230 can download the user's bank customer data and the customer data of other bank customers whose similarity to the user's bank customer data reaches a threshold value from the bank's internal customer data server through the communication interface 130 of device 100, and use a collaborative filtering algorithm to generate the user's potential needs, that is, generate the products that the user may purchase.

[0044] After the demand analysis module 230 of device 100 generates potential user needs, the weight adjustment module 240 of device 100 can set the weight value of mainstream products based on the demand similarity between the demanded products in the user's potential needs and the mainstream products in the consumption trend information generated by the trend analysis module 220 of device 100 (step 350). In this embodiment, it is assumed that the weight adjustment module 240 can first set the weight value of all mainstream products to a predetermined value, and can increase the weight value of mainstream products whose similarity with the demanded products included in the potential needs reaches the demand threshold value, and can also decrease the weight value of mainstream products whose similarity with the demanded products included in the potential needs is lower than a certain value.

[0045] After the weight adjustment module 240 of device 100 sets the weight values ​​of mainstream products, the model maintenance module 250 of device 100 can provide the product information of mainstream products in the consumer trend information generated by the trend analysis module 220 of device 100 and the weight values ​​of mainstream products set by the weight adjustment module 240 to the generative language model (step 350). In this embodiment, it is assumed that the model maintenance module 250 can provide the product information and weight values ​​of mainstream products to the trained generative language model through structured data or natural language instructions.

[0046] After the model maintenance module 250 of device 100 provides the product information and weight values ​​of mainstream products to the generative language model, the real-time interaction module 260 of device 100 can use the generative language model to generate dialogue messages and transmit the generated dialogue messages to client 101 through the communication interface 130 of device 100. Client 101 can then receive the dialogue messages from device 100 via voice and / or text output to continuously interact with the user. The real-time interaction module 260 can also receive the response messages generated and returned by the user during the interaction through the communication interface 130. The weight adjustment module 240 can... The model maintenance module 250 adjusts the weight values ​​of the mainstream products based on the received response messages. It can then provide the adjusted weight values ​​of the mainstream products to the generative language model to update the weight values ​​of the mainstream products in the generative language model (step 360). After that, the real-time interaction module 260 can use the generative language model with the adjusted weight values ​​of the mainstream products to generate dialogue messages again and send them to the client 101 through the communication interface 130. It can also receive response messages again through the communication interface 130 and the weight adjustment module 240 adjusts the weight values ​​of the mainstream products based on the response messages. This process is repeated continuously.

[0047] When the real-time interaction module 260 of device 100 adjusts the weight values ​​of mainstream products based on the response messages generated by client 101 and uses the generative language model with the adjusted weight values ​​of mainstream products to continuously generate dialogue messages to interact with the user (step 360), the product recommendation module 270 of device 100 can determine recommended products based on the weight values ​​of mainstream products (step 370). In this embodiment, it is assumed that the product recommendation module 270 can determine that a mainstream product is a recommended product when the weight value of a certain mainstream product reaches the recommendation threshold, or it can determine the mainstream product with the highest weight value as the recommended product after interacting with the user for a period of time, and provide the instruction or dialogue of the recommended product to the generative language model, so that the generative language model generates dialogue messages of the recommended product based on the product information of the recommended product, and provides the product information of the recommended product to client 101 through the communication interface 130 of device 100.

[0048] After the product recommendation module 270 of device 100 determines the recommended product, client 101 can recommend the user to purchase the recommended product (step 380). In this embodiment, it is assumed that after receiving the dialog message of the recommended product transmitted by device 100, client 101 can display the dialog message of the recommended product and can use voice to ask the user whether to purchase the recommended product.

[0049] Thus, through this invention, the mainstream trends of current consumer goods can be identified by analyzing recent credit card spending data, and a generative language model can be used to interact with users and recommend products for their purchase.

[0050] In the above embodiments, after the trend analysis module 220 of device 100 generates consumer trend information (step 310), and before the weight adjustment module 240 of device 100 sets the weight values ​​of mainstream products in the consumer trend information (step 340), device 100 can also follow the process shown in "Figure 3C". The data acquisition module 210 of device 100 can obtain popular product information through social media platforms, third-party analysis reports, and search engines (step 321). The trend analysis module 220 can use natural language processing technology to extract popular products from the popular product information (step 325). The trend analysis module 220 can also obtain the product information of the extracted popular products, that is, the obtained product information of the popular products can be added to the generated consumer trend information (step 329).

[0051] Additionally, after client 101 recommends the user to purchase the recommended product (step 380), it can also assist the user in purchasing the recommended product. As shown in the process of "Figure 3D", the data acquisition module 210 of device 100 or the banking application of client 101 can query the consumption amount of the recommended product in one or more sales channels (step 391). After that, client 101 can query the card swiping discounts of the credit cards held by the user in the same bank for each sales channel, and can display the consumption amount and card swiping discounts of the recommended product in each sales channel (step 395), so that the user can operate client 101 to select the sales channel to purchase the recommended product. The product recommendation module 270 of client 101 or device 100 can provide the user with the consumption amount of the selected sales channel to swipe the card to purchase the recommended product in the sales channel selected by the user (step 399).

[0052] In summary, the difference between this invention and prior art lies in the following: after the bank server analyzes the historical card-swiping data of bank customers to generate consumption trend information, it uses a generative language model to generate dialogue messages so that the bank application can interact with the user by outputting dialogue messages through voice or text. Based on the interaction with the user, the weight values ​​of mainstream products in the consumption trend information are adjusted, and based on the adjusted weight values ​​of mainstream products, the mainstream products recommended to the user are determined. This technical means can solve the problem in prior art where the products recommended by the bank to bank customers do not meet the purchasing needs of bank customers, thereby achieving the technical effect of increasing the willingness of bank customers to purchase products recommended by the bank.

[0053] Furthermore, the method of recommending products based on card swipe records and user dialogue in this invention can be implemented in hardware, software, or a combination of hardware and software. It can also be implemented in a centralized manner in a computer system or in a decentralized manner with different components distributed among several interconnected computer systems.

[0054] While the embodiments disclosed in this invention are as described above, the content is not intended to directly limit the scope of patent protection for this invention. Any modifications or refinements made in form and detail to the implementation of this invention by those skilled in the art, without departing from the spirit and scope disclosed herein, shall fall within the scope of patent protection for this invention. The scope of patent protection for this invention shall still be determined by the claims outlined in the appended patent applications.

[0055] 100: Device 101: Client 110: Memory 130: Communication Interface 140: Storage Media 170: Processor 190: Busbar 210: Data Acquisition Module 220: Trend Analysis Module 230: Requirements Analysis Module 240: Weight Adjustment Module 250: Model Maintenance Module 260: Real-time Interaction Module 270: Product Recommendation Module Step 301: The bank server retrieves the bank customer's historical card swipe records. Step 310: The bank server analyzes the historical card-swiping data of bank customers to generate consumption trend information, which includes product information of mainstream goods. Step 311: The bank server uses a clustering analysis model to classify each item in the historical card swipe records. Step 313: The bank server uses a time-series analysis model to analyze the transaction trends of each card-swiping item from historical card-swiping records. Step 315: The bank server uses an association analysis model to find the transaction correlation between the combinations of goods and / or consumption behaviors that appear a certain number of times in historical card swipe records and the goods swiped. Step 317: The bank server calculates the overall index score for each card-swipe item's classification, transaction trend, item combination, and transaction correlation based on the weights of the cluster analysis model, time series analysis model, and association analysis model. Step 319: The bank server sorts each card-swiping item according to the overall index score to generate consumption trend information. Step 321: The bank's server obtains information on popular products through social media platforms, third-party analytics reports, and search engines. Step 325: The bank server uses a natural language processing model to extract popular products from the popular product information. Step 329: The bank server adds information on popular products to the consumer trend information. Step 330: The bank server analyzes the user's bank customer data to generate potential demand. Step 340: The bank server assigns a weight to mainstream products based on the similarity between the products in potential demand and mainstream products. Step 350: The bank server provides product information and weight values ​​of mainstream products to the generative language model. Step 360: The bank server uses a generative language model to generate dialogue messages, enabling the bank application to continuously interact with the user by outputting these messages in language and / or text. Furthermore, the bank server adjusts the weight values ​​of mainstream products based on the user's responses during the interaction and updates the adjusted weight values ​​of the mainstream products to the generative language model. Step 370: The bank server determines recommended products based on the weight values ​​of mainstream products. Step 380: The bank app recommends products to users. Step 391: Use the bank's server or application to access the sales channel and check the purchase amount for recommended products. Step 395: The bank app displays the spending amount and credit card discounts from the sales channel. Step 399: The bank server or bank application allows customers to purchase recommended products by swiping their cards using the amount spent through the sales channel.

Claims

1. A method for recommending products based on credit card transaction records and user interaction, the method comprising at least the following steps: A bank server obtains historical credit card transaction records of multiple bank customers; The bank server analyzes the historical credit card transaction data of the bank customers using at least one computational model to generate a consumption trend message, the consumption trend message including product information of multiple mainstream products selected from a variety of credit card transaction products; The bank server analyzes the bank customer data of a user to generate a potential demand, the user being one of the bank customers, and the potential demand including at least one desired product; The bank server assigns weight values ​​to the mainstream products based on the similarity of the at least one desired product to the mainstream products; The bank server provides the product information and weight values ​​of the mainstream products to a generative language model, the generative language model being trained using a pre-established interactive paradigm; The bank server uses a generative language model to generate multiple dialogue messages and transmits these messages to a banking application. The banking application then outputs these dialogue messages via voice and / or text to continuously interact with the user. During the interaction, the bank server adjusts the weight of at least one of the mainstream products based on the semantic tendency of the user's response (whether it's a positive response to a mainstream product or a lack of response to any mainstream product or an opinion or rating of a mainstream product). This adjusted weight is then provided to the generative language model, which generates the dialogue messages based on these weights. The bank server determines a recommended product based on the weights of the mainstream products, and the banking application recommends that the user purchase the recommended product.

2. The method for recommending products based on credit card swipe records and user dialogue as described in claim 1, wherein the bank server uses at least one computational model to analyze the historical credit card swipe data to generate the consumption trend information, further comprising the steps of the bank server using a clustering analysis model to classify each credit card swipe item in the historical credit card swipe records according to the transaction merchant, consumption behavior, investment preference, transaction time, or transaction location; using a time series analysis model to analyze the transaction trend of each credit card swipe item according to the transaction time and / or transaction location in the historical credit card swipe records; using a correlation analysis model to find combinations of credit card swipe items that appear a certain number of times in the historical credit card swipe records and / or the transaction correlation between consumption behavior and one of the credit card swipe items; and calculating a total index score of the classification of the credit card swipe items, the transaction trend, the product combination, and the transaction correlation according to the weights of the clustering analysis model, the time series analysis model, and the correlation analysis model; and sorting each credit card swipe item according to the total index score to generate the consumption trend information.

3. The method for recommending products based on card swipe records and conversations with users as described in claim 1, wherein the bank server uses at least one computational model to analyze the historical card swipe data to generate the consumption trend information, further includes the steps of the bank server obtaining a message of interest through social media platforms, third-party analysis reports, and / or search engines, extracting at least one popular product from the message of interest using a natural language processing model, and adding the product information of the at least one popular product to the consumption trend information.

4. The method for recommending products based on card swipe records and user conversations as described in Request 1, wherein the step of the bank server generating the consumption trend information further includes calculating the index scores of each card-swipe product in terms of category, transaction trend, product combination, and transaction correlation according to predefined weights, summing the index scores of each card-swipe product to generate a total index score, and selecting the mainstream products based on the total index scores of each card-swipe product.

5. The method for recommending products to a user based on a card swipe record as described in claim 1, wherein after the step of the bank application recommending the user to purchase the recommended product, the method further includes the bank server or the bank application querying the consumption amount of the recommended product through at least one sales channel, and the bank application displaying the consumption amount and card swipe discount of the at least one sales channel, and the bank application providing the step of using the consumption amount of the sales channel to swipe the card to purchase the recommended product through the sales channel.

6. An apparatus for recommending products based on card swipe records and user interaction, providing a client connection, the apparatus comprising at least: a data acquisition module for acquiring historical card swipe records of multiple bank customers; a trend analysis module for analyzing the historical card swipe data of the bank customers using at least one computational model to generate a consumption trend message, the consumption trend message including product information of multiple mainstream products selected from a variety of card swipe products; a demand analysis module for analyzing the bank customer data of a user to generate a potential demand, the user being one of the bank customers, and the potential demand including at least one desired product; a weight adjustment module for setting weight values ​​for the mainstream products based on the similarity of the at least one desired product to the mainstream products; and a model maintenance module for providing product information and weight values ​​of the mainstream products to a generative language model, the generative language model being trained using a pre-established interactive paradigm; An instant interaction module is used to generate multiple dialogue messages using a generative language model and transmit these dialogue messages to the client, enabling the client to continuously interact with the user by outputting these dialogue messages via voice and / or text. A weight adjustment module is further used to adjust the weight value of at least one of the mainstream products based on the semantic tendency of the user's response during the interaction—whether it is a positive response to a mainstream product or a lack of response to any mainstream product or an opinion or rating of a mainstream product. The model maintenance module provides the adjusted weight values ​​of the mainstream products to the generative language model, enabling the generative language model to generate the dialogue messages based on these adjusted weight values. A product recommendation module is also included, used to determine a recommended product based on the weight values ​​of the mainstream products, allowing the client to recommend the recommended product to the user.

7. The device for recommending products to users based on credit card swipe records as described in claim 6, wherein the trend analysis module is further used to classify each credit card swipe item in historical credit card swipe records using a clustering analysis model based on the transaction merchant, consumption behavior, investment preference, transaction time, or transaction location; analyze the transaction trend of each credit card swipe item using a time series analysis model based on the transaction time and / or transaction location in historical credit card swipe records; use a correlation analysis model to find product combinations and / or consumption behaviors that appear a certain number of times in the historical credit card swipe records and their transaction correlation with the credit card swipe items; calculate a total index score for the classification, transaction trend, product combination, and transaction correlation of the credit card swipe items based on the weights of the clustering analysis model, the time series analysis model, and the correlation analysis model; and sort each credit card swipe item according to the total index score to generate the consumption trend information.

8. The device for recommending products based on card swipe records and conversations with users as described in claim 6, wherein the data acquisition module is further used to obtain a message of interest through social media platforms, third-party analysis reports, and / or search engines, and the trend analysis module is further used to extract at least one popular product from the message of interest using natural language processing models and topic models, and to add the product information of the at least one popular product to the consumer trend message.

9. The device for recommending products based on card swipe records and user dialogue as described in claim 6, wherein the trend analysis module calculates the index scores of each card-swipe product in terms of category, transaction trend, product combination, and transaction correlation according to predefined weights, sums up the index scores of each card-swipe product to generate a total index score, and selects the mainstream products based on the total index scores of each card-swipe product.

10. The device for recommending products to a user based on a card swipe record as described in claim 6, wherein the product recommendation module is further configured to connect to at least one sales channel to query the consumption amount of the recommended product, and provide the option to purchase the recommended product by swiping a card in the at least one sales channel using the consumption amount of the at least one sales channel.

Citation Information

Patent Citations

  • Credit card preferential information recommendation method and system based on user consumption portrait

    CN117745348A

  • Management method, system and equipment integrating large language model and voice recognition

    CN118350859A

  • Call service processing method, system and device, storage medium and program product

    CN118450053A

  • Method and system of personalized discount recommendation relating to a marketing method with a personalized discount recommendation system

    TW202437172A

  • Personalized discount recommendation methods and systems

    TWI847588B