Device and method for interaction and self-adjustment according to multi-platform data prediction behaviors

Abstract

A device and method for performing interaction and self-adjustment according to multi-platform data prediction behaviors are provided, which analyzes the interaction behavior data of a user on a plurality of platforms, predicts the behavior type of the user by using an artificial intelligence model according to the analysis result, and generates personalized response information according to the predicted behavior type to interact with the user. And the technical means of adjusting the artificial intelligence model according to the interactive behavior of the user can accurately grasp the intention of the user in a cross-platform manner and enable the model to be adaptively adjusted, and the technical effects of accurately interacting with the user in a personalized manner and improving the service efficiency are achieved.

Description

TECHNICAL FIELD

[0001] A user interaction device and method thereof, in particular, a device and method for interacting and self-adjusting based on multi-platform data prediction behavior. BACKGROUND

[0002] With the popularity of digital services and social media, user behavior data is scattered across different platforms, such as posts and comments on social media platforms, shopping records on e-commerce platforms, instant messaging and conversations with customer service, etc.

[0003] In current behavior analysis techniques, data is usually analyzed from a single platform or a single source, such as analyzing browsing records and purchase records on e-commerce platforms to push personalized product information, such as product advertisements that users are interested in. In this way, although personalized recommendation information can be provided on the platform, it is not possible to fully capture the user's behavior patterns across different platforms. For example, a user may express a shopping intention on a social media platform, but has not completed a transaction on an e-commerce website. Using data from a single platform often cannot link the two, resulting in insufficient accuracy of behavior prediction, making the effect of personalized information unsatisfactory.

[0004] In addition, existing behavior prediction techniques mostly rely on static data and lack the ability to make immediate judgments about user emotions. Even some behavior prediction techniques use natural language processing (NLP) techniques to analyze user posts or comment content, but are limited to sentiment polarity (positive or negative) classification and do not capture changes in user emotional tendencies in different situations, which also limits the effectiveness of personalized response information.

[0005] Furthermore, although some behavior prediction techniques currently incorporate machine learning models, the models often lack mechanisms for continuous updating and adjustment and cannot make immediate corrections to the machine learning model based on response effectiveness, which can cause the prediction results to gradually deviate from actual behavior.

[0006] In summary, it can be seen that the existing technology has long been plagued by the problem of ineffective personalized information generated by online platforms due to the lack of consideration of emotional tendencies and the lack of immediate adjustments. Therefore, it is necessary to propose improved technical means to solve this problem. SUMMARY

[0007] In view of the problem of ineffective personalized information generated by online platforms due to the lack of consideration of emotional tendencies and the lack of immediate adjustments in the existing technology, the present application discloses a device and method for interacting and self-adjusting based on multi-platform data prediction behavior, wherein:

[0008] The disclosed device for interacting with and self-adjusting to predicted behavior based on multi-platform data comprises at least: a data transmission module for obtaining user interaction behavior data from multiple data sources and for obtaining user historical interaction data; a data integration module for integrating the interaction behavior data; a feature analysis module for analyzing the interaction behavior data to generate a user feature vector and an emotional tendency index; a behavior prediction module for inputting the user feature vector and the emotional tendency index as parameters into an artificial intelligence model, causing the artificial intelligence model to generate a predicted behavior type of the user on a target platform; a response generation module for generating personalized response information for the user on each target platform according to the predicted behavior type; a response monitoring module for monitoring the user interaction behavior of the user after providing the personalized response information to the user to generate a monitoring result; and a model adjustment module for adjusting parameters in the artificial intelligence model related to generating the predicted behavior type according to the monitoring result.

[0009] The disclosed method for interacting with and self-adjusting to predicted behavior based on multi-platform data comprises at least the following steps: obtaining user interaction behavior data from multiple data sources; integrating the interaction behavior data; analyzing the interaction behavior data to generate a user feature vector and an emotional tendency index; inputting the user feature vector and the emotional tendency index as parameters into an artificial intelligence model, causing the artificial intelligence model to generate a predicted behavior type of the user on a target platform; generating personalized response information for the user on each target platform according to the predicted behavior type; monitoring the user interaction behavior of the user after providing the personalized response information to the user to generate a monitoring result; and adjusting parameters in the artificial intelligence model related to generating the predicted behavior type according to the monitoring result.

[0010] The disclosed device and method differ from the prior art in that the present invention analyzes user interaction behavior data on multiple platforms, uses an artificial intelligence model to predict the possible behavior type of the user according to the analysis results, generates personalized response information according to the predicted behavior type, and transmits the response information to interact with the user, and can adjust the artificial intelligence model according to the user interaction behavior, thereby solving the problems existing in the prior art and achieving the technical effects of precise and personalized interaction with the user and improving service efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 Element schematic diagram of the disclosed device for interacting with and self-adjusting to predicted behavior based on multi-platform data.

[0012] Figure 2 Module schematic diagram of the processor.

[0013] Figure 3AMethod flowchart for the proposed method of interacting and self-adjusting the behavior of the data prediction platform.

[0014] Figure 3B Method flowchart for the proposed method of generating personalized response information for the user.

[0015] Figure 3C Method flowchart for the proposed method of selecting the optimal interaction channel to deliver personalized response information.

[0016] Figure 3D Method flowchart for the proposed method of determining key features to update the artificial intelligence model.

[0017] Figure 3E Method flowchart for the proposed method of determining negative experiences and performing remedies.

[0018] Figure 4 Schematic diagram of the proposed method of obtaining interaction behavior data from multiple data sources for analysis and providing personalized response information.

[0019] Explanation of reference signs:

[0020] 100: device

[0021] 110: memory

[0022] 130: communication interface

[0023] 140: storage medium

[0024] 170: processor

[0025] 190: bus

[0026] 210: data transmission module

[0027] 220: data integration module

[0028] 230: feature analysis module

[0029] 240: behavior prediction module

[0030] 250: response generation module

[0031] 260: response monitoring module

[0032] 270: model adjustment module

[0033] 275: artificial intelligence model

[0034] 411: social media platform

[0035] 412: e-commerce website

[0036] 413: Official website of the enterprise

[0037] 421, 422: Personalized response information

[0038] Step 310: Obtain interaction behavior data of the user in multiple data sources

[0039] Step 320: Integrate the interaction behavior data

[0040] Step 330: Analyze the interaction behavior data to generate a user feature vector and an emotional tendency index

[0041] Step 340: Input the user feature vector and the emotional tendency index as parameters into an artificial intelligence model, so that the artificial intelligence model generates a predicted behavior type of the user on a target platform

[0042] Step 350: Generate personalized response information of the user on the target platform according to the predicted behavior type

[0043] Step 351: Determine the current situation of the user

[0044] Step 353: Analyze the user's public posts on social media in the interaction behavior data by natural language processing technology to generate a mirror communication style

[0045] Step 355: Generate personalized response information in accordance with the mirror communication style using corresponding interaction strategies according to the emotional tendency index, the current situation, the predicted behavior type, and historical interaction data

[0046] Step 361: Select the best interaction channel according to the predicted behavior type, the current situation, and the user's preferences

[0047] Step 365: Provide personalized response information to the user using the best interaction channel

[0048] Step 370: Monitor the user's interaction behavior after providing the personalized response information to the user to generate a monitoring result

[0049] Step 380: Adjust the parameters in the artificial intelligence model related to generating the predicted behavior type according to the monitoring result

[0050] Step 381: Perform attribution analysis on the monitoring result to determine the key features that affect the user's interaction behavior

[0051] Step 385: Update the weights in the artificial intelligence model corresponding to the key features DETAILED DESCRIPTION

[0052] 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 any person 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.

[0053] This invention can analyze user interaction data across multiple platforms, use an artificial intelligence model to predict possible user behavior types based on the analysis results, generate personalized response information based on the predicted behavior types, and adjust the artificial intelligence model based on the user's interaction behavior after receiving the personalized response information.

[0054] The apparatus for implementing this invention can be a computing device. The computing device of this invention 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 the included 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 disposed.

[0055] The bus of the computing device proposed in this invention can 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) bus, Peripheral Component Interconnect (PCI) bus, Video Electronics Standards Association (VESA) local bus, and serialized Universal Serial Bus (USB), PCI Express (PCI-E / PCIe) bus, etc.

[0056] 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 may interpret a computer instruction or a series of multiple computer instructions to perform specific operations or calculations, 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., as well as conventional procedural programming languages ​​such as C or other similar programming languages.

[0057] 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 a bus. A chipset consists of one or more integrated circuits (ICs), including a memory controller and peripheral input / output (I / O) controllers, etc. That is, the memory controller and I / O controllers 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 special-purpose registers, timers, etc., which can be accessed or used by one or more processing modules coupled to or electrically connected to the chipset. In some embodiments, the chipset may also be part of the processing module.

[0058] 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), Flash memory, etc. The aforementioned mass storage areas can contain any type of storage device or storage medium, such as hard disk drives, optical discs, flash drives, memory cards, solid-state drives (SSDs), or any other storage device. In other words, the memory controller can access data in static random access memory, dynamic random access memory, flash memory, hard disk drives, and solid-state drives.

[0059] 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, or joystick. Peripheral output devices can be any type of output device, such as a monitor or printer. Peripheral input devices and peripheral output devices can also be the same device, such as a touchscreen. Communication interfaces can include wireless communication interfaces 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 and other 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. The data or signal receiving device may include a GPS receiver or a physiological signal receiver, the physiological signals received by the physiological signal receiver including, but not limited to, heartbeat, blood oxygenation, etc. The processing module may periodically poll various peripheral devices and interfaces, enabling the computing device to input and output data through various peripheral devices and interfaces, and also to communicate with another computing device having the hardware components described above.

[0060] The following is a preliminary step. Figure 1 The schematic diagram of the device for interactive and self-adjusting based on multi-platform data prediction behavior proposed in this invention illustrates the device for implementing this invention. For example... Figure 1 As shown, 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 memory 110, the communication interface 130, the storage medium 140, and the processor 170 are interconnected via the bus 190.

[0061] The memory 110 can store one or more sets of computer instructions.

[0062] The communication interface 130 can be connected to external network storage devices or servers, and request and download data from the connected network devices.

[0063] Storage medium 140 can store data or signals downloaded from 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.

[0064] Processor 170 can be like Figure 2 The schematic diagram of the modules proposed in this invention shows that they include modules such as a data transmission module 210, a data integration module 220, a feature analysis module 230, a behavior prediction module 240, a response generation module 250, a response monitoring module 260, and a model adjustment module 270. In some embodiments, the processor 170 can execute computer instructions stored in the memory 110, and can generate a response after executing the computer instructions. Figure 2 The various modules within; in another embodiment, Figure 2 The modules within can be generated from one or more circuits and / or complete or partial hardware components such as chips; that is, the processor 170 comprises... Figure 2 The hardware components of each module in the processor 170, that is, each module included in the processor 170 can be a software module or a hardware module, and there are no particular limitations in this invention.

[0065] The data transmission module 210 is responsible for acquiring user interaction behavior data from multiple data sources and storing the acquired interaction behavior data in the storage medium 140. The data transmission module 210 can acquire interaction behavior data from data sources such as social media platforms, e-commerce websites, and corporate websites through the communication interface 130. More specifically, the interaction behavior data can be, as obtained by the data transmission module 210 from social media platforms, user-posted article content, user click records, user browsing history, user comments or response emoticons, user likes or shares, etc.; it can also be, as obtained by the data transmission module 210 from e-commerce platforms, user browsing records, user shopping records, and payment records; or it can be, as obtained by the data transmission module 210 from corporate websites, information on communication between users and customer service, etc. However, the interaction behavior data mentioned in this invention is not limited to the above.

[0066] The data transmission module 210 can also provide users with ways to participate in online voting, opinion surveys, website activities, etc., to encourage users to actively provide zero-party data such as user preferences, needs or intentions, and can add the zero-party data received through the communication interface 130 to the interactive behavior data.

[0067] The data transmission module 210 can also obtain the user's historical interaction data. For example, it can download the user's historical interaction data by connecting to a network device that stores historical interaction data through the communication interface 130, or directly read the user's historical interaction data that has been stored in the storage medium 140. The historical interaction data obtained by the data transmission module 210 includes the user's current interactive behavior after receiving the personalized response information generated by the response generation module 250, etc., but the present invention is not limited thereto.

[0068] The data transmission module 210 is also responsible for providing the personalized response information generated by the response generation module 250 to the user. Specifically, the data transmission module 210 can directly provide the personalized response information to the user. For example, the data transmission module 210 can provide personalized response information to the user through the communication interface 130 using the channel selected by the behavior prediction module 240. These channels include email, instant messaging, application push notifications, and SMS messages, but the invention is not limited to these. Alternatively, the data transmission module 210 can indirectly provide personalized response information to the user through a data source. For example, the data transmission module 210 can transmit the personalized response information to the data source through the communication interface 130, allowing the data source to provide the personalized response information to the user through internal advertising or internal related recommendation information. For instance, when the data source is a social media platform, the related recommendation information could be articles or videos recommended to the user; when the data source is an e-commerce website, the related recommendation information could be products recommended to the user.

[0069] The data integration module 220 is responsible for integrating the interactive behavior data obtained by the data transmission module 210. For example, the data integration module 220 can perform format standardization conversion on interactive behavior data from multiple data sources with different data formats, so that interactive behavior data from different data sources have the same format and structure after format conversion.

[0070] The data integration module 220 can also perform de-identification processing on the interactive behavior data in accordance with the principle of privacy protection. For example, it can perform hash operations on sensitive data (such as email accounts, phone numbers, etc.) in the interactive behavior data, or use anonymization or differential privacy technology to remove personally identifiable information in the interactive behavior data. However, the present invention is not limited to this.

[0071] The data integration module 220 can also construct and maintain an identity graph to associate user interaction data from different data sources, thereby allowing interaction data to be mapped to the same user across platforms. An identity graph is typically represented by a graph structure from graph theory in discrete mathematics. In the identity graph, each node represents a user, and each edge represents the association between the user and the data source. The data integration module 220 can merge nodes (representing associating users from different platforms as the same person) or update edge weights each time the identity graph is maintained, to gradually improve the accuracy of identifying the same user.The data integration module 220 can maintain the identity graph through deterministic matching and probabilistic matching. More specifically, when data from different sources contains clearly corresponding identification data (such as login accounts, linked email addresses, phone numbers, or member identification data), the data integration module 220 uses a deterministic matching algorithm to associate data from different sources that correspond to the same identification data with nodes representing the same user in the identity graph. For example, if a user registers on a social media platform with a certain email address and also uses the same email address to create a membership account on an e-commerce website, the data integration module 220 can merge the user's interaction behavior data on the social media platform registered with that email address with the interaction behavior data on the e-commerce website where the membership account was created, into interaction behavior data associated with that email address. Furthermore, the data integration module 220 can even integrate internal customer relationship management (CRM) data. The data integration module 220 can also correlate customer data and interaction behavior data in a CRM (Customer Relationship Management) system. When data from different sources does not contain clearly corresponding identification data, the module can use probabilistic association algorithms to infer whether certain users from different data sources belong to the same person, based on device characteristics (such as browser fingerprints, network address ranges), behavioral patterns (such as frequent usage time periods, click habits), language preferences, or geographical location, to update the identity graph. For example, if an anonymous visitor on a social media platform and a social media platform account exhibit highly similar browsing behavior using the same device type client during the same time period, the data integration module 220 can determine that the anonymous visitor and the social media platform account are the same user. The probability is relatively high; at the same time, the data integration module 220 can periodically improve the accuracy of the association (i.e., identity graph) of users from different data sources through verification loops. For example, the data integration module 220 can perform consistency checks at regular intervals. For instance, when users from different data sources that are associated exhibit contradictory behaviors in their respective data sources (such as setting the age range to 18-25 on one platform and setting the age to 50 and above on another platform), the data integration module 220 can mark the nodes in the identity graph that represent users from different data sources as having reduced reliability of association. In addition, the data integration module 220 can even conduct simple interactions with users from different data sources (such as preference surveys or login verification) to verify the accuracy of the associated users.For a practical example, if a user registers with a certain email address on a social media platform and leaves a message using a nickname, registers with the same email address on an e-commerce website and purchases goods, and leaves a device fingerprint on the customer service platform of a company's official website, then the data integration module 220 can use deterministic matching (email address), probabilistic matching (device characteristics), and verification loops (consistent language and behavioral habits across platforms) to associate the user's interactive behavior data on the three social media platforms, the e-commerce website, and the company's official website with the same nodes in the identity graph.

[0072] The feature analysis module 230 is responsible for analyzing the interactive behavior data integrated by the data integration module 220 to generate user feature vectors. For example, the feature analysis module 230 can first convert the interactive behavior data into quantifiable feature vectors. More specifically, the feature analysis module 230 can use word embedding models such as Word2Vec and BERT to convert user-related topics in text data (such as article content, comments, and communication with customer service) into text vectors. The feature analysis module 230 can also perform time-series analysis on behavioral data (such as clicks, views, and purchases) and convert it into numerical features, such as "number of purchases = 5" and "stay time = 120 seconds". The feature analysis module 230 can also use Mel-Frequency Cepstral Coefficient (MFCC) and prosody theory to identify speech intonation and convert speech signals into acoustic features. Furthermore, it can use convolutional neural networks (CNNs). A network (CNN) extracts and identifies facial expressions in image signals to generate image features. In some embodiments, the feature analysis module 230 can also perform normalization transformation on the interactive behavior data, so that the numerical range of the interactive behavior data falls within a consistent range after transformation. For example, the numerical range can be transformed to 0-1 (e.g., a dwell time of 120 seconds becomes 0.8 after normalization, and the number of purchases of 5 becomes 0.4 after normalization), or the numerical range can be transformed to a standard normal distribution. Next, the feature analysis module 230 can concatenate the generated text vectors, numerical features, voice features, and image features to combine them into a high-dimensional user feature vector.

[0073] The feature analysis module 230 is responsible for analyzing the interactive behavior data integrated by the data integration module 220 to generate sentiment tendency indicators. For example, the feature analysis module 230 can use natural language processing (NLP) techniques such as BERT and RoBERTa to analyze the sentiment polarity (e.g., positive, neutral, negative) and emotion category (e.g., anxiety, joy, anger) of text data, and can use convolutional neural networks to detect facial key points and expression features in image signals to determine the emotion category; the feature analysis module 230 can also analyze behavioral patterns in interactive behavior data by analyzing dwell time and interaction frequency, mining behavioral sequences, and detecting interaction tendencies and preference differences, and can use techniques such as Support Vector Machine (SVM), random forest, or deep neural networks to analyze behavioral patterns in interactive behavior data. Emotion classification models such as Dependent Neural Networks (DNNs) infer emotion categories. For example, calculating the time a user spends on a particular content or page and the number of interactions indicates high interest and positive emotions when they spend a long time and interact frequently by clicking and leaving comments, while short stays and quick exits indicate indifference or negative emotions. If a user's interaction behavior data shows "clicking an ad → immediately closing it → never returning," the emotion category is aversion / rejection. If the user's interaction behavior data shows "browsing a product page → adding it to the cart → hesitating → not checking out," the emotion category is anxiety / consideration. If a user has a long-term tendency to "actively reply to comments" but recently shows "silence and ignoring information," it may represent negative emotions. If there is a sudden and significant increase in interaction behavior recently, it may represent excitement or positive emotions. The feature analysis module 230 can also acquire the fundamental frequency (F0), analyze the speech rate, and monitor volume changes of the speech signal to obtain speech features such as fundamental frequency, speech rate, and volume changes. Then, it can use convolutional neural networks or recurrent neural networks to obtain these speech features. A Recurrent Neural Network (RNN) analyzes speech features to determine emotion categories. The feature analysis module 230 can analyze the speech signal using a Short-time Fourier Transform (STFT) or Autocorrelation Function (Autocorrelation Function) and then detect the vocal cord vibration period to obtain the fundamental frequency of the speech signal. The feature analysis module 230 can also segment the speech signal into words or syllables using energy thresholding or endpoint detection to calculate the average number of syllables or words per second to estimate the speech rate. Furthermore, the feature analysis module 230 can calculate the energy variation amplitude of each short time window within the speech signal to detect volume increases, decreases, or stable intervals, thereby monitoring volume changes in the speech signal.Then, the feature analysis module 230 can also fuse the emotion categories determined based on text data, voice features, image features, and behavioral data to generate a set of overall emotion tendency indicators. Specifically, the feature analysis module 230 can input the emotion results determined based on text data, voice features, image features, and behavioral data into a deep learning model to generate emotion tendency indicators, or it can perform weighted calculations on the emotion results determined based on text data, voice features, image features, and behavioral data according to different weights to generate emotion tendency indicators.

[0074] In some embodiments, the feature analysis module 230 can establish a long-term and short-term interest decay model and use the long-term and short-term interest decay model to divide the user feature vector into a long-term interest vector and a short-term intent vector. The long-term interest decay model uses a time decay function to give a higher weight to short-term behavior. For example, when analyzing the interaction behavior data of a user who has liked photography for a long time but has recently been frequently searching for camera discounts, the long-term and short-term interest decay model gives a larger proportion of short-term intent to the feature vector.

[0075] The feature analysis module 230 can also calculate emotional volatility indicators or contrarian psychological indicators based on the historical sequence of the user's emotional tendency indicators. This allows the module to measure the user's sensitivity and reaction tendency to the personalized response information generated by the response generation module 250. This enables the AI ​​model used by the behavior prediction module 240 to correct subsequent behavior predictions and allows the response generation module to adjust its personalized response information generation strategy. More specifically, the feature analysis module 230 can acquire the user's emotional tendency value (e.g., emotional scores from -1 to +1, representing negative to positive emotions) at regular intervals (different time periods) when the user interacts with the data source. It can also calculate the user's order difference sequence based on the changes in emotional scores over consecutive time periods. Furthermore, it can generate an emotional volatility indicator based on the standard deviation or mean absolute change of the calculated difference sequence. A higher value indicates greater emotional fluctuation, while a lower value indicates more stable emotions. The feature analysis module 230 can also obtain the user's interactive behavior after receiving personalized response information from the data source, especially negative behaviors (e.g., immediately closing push notifications). The feature analysis module 230 can compare the "actual emotional reaction" of user interaction behavior with the "expected emotional effect" when generating personalized response information. If the actual emotional reaction is opposite to the expected emotional effect (such as the user's emotion is still negative after the reassurance information), the feature analysis module 230 can increase the resistance score for that interaction. Afterwards, the feature analysis module 230 can calculate the resistance psychological index based on the cumulative frequency and intensity of resistance events over a period of time. For example, if the user's emotional tendency changes from neutral to negative when receiving promotional information multiple times, the feature analysis module 230 can determine that the user has a high degree of resistance to the promotional information, and the value of its resistance psychological index will be increased accordingly.

[0076] The behavior prediction module 240 is responsible for providing the user feature vector and sentiment index generated by the feature analysis module 230 as parameters to the artificial intelligence model maintained by the model adjustment module 270. This allows the artificial intelligence model to generate predicted behavior types for the user on the corresponding target platform based on the user feature vector and sentiment index. The behavior prediction module 240 can generate a single predicted behavior type applicable to all target platforms, a predicted behavior type for a subset of target platforms, or a predicted behavior type for each target platform individually. The artificial intelligence model can be a sequence behavior prediction model using sequence behavior prediction techniques, such as a recursive neural network (RNN), a long short-term memory (LSTM) model, or a Transformer model. By analyzing the time series of user feature vectors and sentiment indexes, it can predict the user's next action on each target platform. The target platform can be any data source. For example, the artificial intelligence model can predict that the user might click on a specific product or turn off marketing notifications on an e-commerce platform, post product recommendation articles or leave comments recommending products or browsing product-related articles on a social media platform, or ask customer service questions about products on a company's official website.

[0077] The behavior prediction module 240 can also determine the user's current situation based on interaction behavior data. For example, the current situation can correspond to the Customer Journey model, which can be divided into five stages: awareness, consideration, purchase, service, and loyalty. If a user is watching or using a search engine to search for product information on social media platforms, such as watching articles or videos related to mobile phone introductions on social media platforms, searching for "mobile phone recommendations" on search websites, or continuously browsing product pages on mobile phone brand websites, the behavior prediction module 240 can determine that the user's current situation is in the awareness stage. If the user starts comparing different products or specifications, such as repeatedly browsing multiple camera product pages of different brands and specifications on e-commerce platforms within a short period of time, adding some camera products to the shopping cart but not yet checking out, or posting on social media platforms asking for advice, the behavior prediction module 240 can determine that the user's current situation is in the consideration stage. If the user enters the actual transaction process, such as interacting with... If the behavioral data records that a user enters the checkout page and checks payment methods, inquires about discounts in conversations with customer service, or attempts to enter a discount code on the checkout page, the behavior prediction module 240 can determine that the user's current situation is in the purchase stage. If the user needs installation, after-sales service, or problem-solving assistance, for example, if the interactive behavior data records conversations between the user and customer service or data uploaded by the user on the product service webpage containing phrases such as "cannot be turned on," "needs warranty service," or "what to do if it breaks down," the behavior prediction module 240 can determine that the user's current situation is in the service stage. If the user shows continuous support or recommendation behavior for a specific brand or product, for example, if the interactive behavior data records positive reviews of the brand or product on social media platforms (e.g., "will buy again next time"), participation in membership activities, or repurchase of the same brand's products on e-commerce platforms, the behavior prediction module 240 can determine that the user's current situation is in the loyalty stage. Furthermore, the behavior prediction module 240 can use the user's historical interaction data obtained by the data transmission module 210 to improve the accuracy of judging the current situation. For example, when a user searches for "gift" related keywords on a shopping platform before a holiday, the behavior prediction module 240 can determine that the user's current situation is in the cognitive stage.

[0078] The behavior prediction module 240 can select the channel most likely to elicit a positive user response based on the predicted behavior type, the current context, and the user preferences obtained by the data transmission module 210. In this invention, the channel selected by the behavior prediction module 240 is also referred to as the "optimal interaction channel." For example, when a user frequently receives high-value promotional information via push notifications, the behavior prediction module 240 can select push notifications as the optimal interaction channel for high-value promotional information.

[0079] The behavior prediction module 240 can also perform contextual analysis on interactive behavior data to determine user intent, thereby correcting the interpretation of the same input information based on the sequence of interactive behavior data. More specifically, the behavior prediction module 240 can determine user intent based on interactive behavior data such as user search history, browsing history, items added to shopping cart, and click behavior on e-commerce platforms, as well as interactive behavior data such as user posts or comments on social media platforms. For example, when a user browses Q&A on a customer service page, the system can infer that their intent is "seeking assistance" rather than "shopping"; and if a user searches for "running shoes" on an e-commerce platform and repeatedly browses products categorized as "running shoes," and posts an article on a social media platform stating "ready to start exercising," the behavior prediction module 240 can determine that the user's intent may be related to exercise.

[0080] The response generation module 250 is responsible for generating personalized response information for each data platform based on the predicted behavior type generated by the behavior prediction module 240. In some embodiments, the response generation module 250 can also incorporate historical interaction data obtained by the data transmission module 210 to generate personalized response information. For example, the response generation module 250 can first obtain the interaction strategies corresponding to different predicted behavior types. For instance, on an e-commerce website, the strategy corresponding to the predicted behavior type of purchasing is to provide the user with an immediate checkout discount or limited-time promotional information; on a corporate website, the strategy corresponding to the predicted behavior type of contacting customer service is to notify the user that customer service will proactively contact them. The response generation module 250 can provide the obtained interaction strategies to a Large Language Model (LLM) and use the LLM to generate personalized response information that matches the obtained interaction strategies. It should be noted that the response generation module 250 can also add an information identification code corresponding to the user and the predicted behavior type to the personalized response information generated by the LLM, for example, by adding an information identification code to the link of the personalized response information, but this invention is not limited thereto.

[0081] The personalized response information generated by the response generation module 250 can be either immediate or delayed. For example, an immediate response can be given first via push notification or instant messaging, followed by a delayed response via email.

[0082] In some implementations, the response generation module 250 can analyze the user's publicly posted articles on social media platforms from the interaction behavior data obtained by the data transmission module 210 using natural language processing technology to learn the user's language style. This allows it to generate personalized response messages that closely match the user's tone and manner of speaking, further enhancing the interaction effect. In this invention, the user's language style learned by the response generation module 250 is also referred to as a "mirror communication style."

[0083] The response generation module 250 can also generate personalized response information that matches the generated mirrored communication style by using corresponding interaction strategies based on the emotion tendency index generated by the feature analysis module 230, the predicted behavior type generated by the behavior prediction module 240, the current situation, and the historical interaction data obtained by the data transmission module 210. The interaction strategies can be predefined in an interaction strategy database established based on historical interaction data. This database records strategy rules corresponding to different current situations, emotion tendencies, and predicted behavior types. Interaction strategies can also be generated based on the context of the current situation and the interaction behavior data. For example, when the user is in the consideration stage, the interaction strategy can focus on discounts or product comparisons; while when the user is in the service stage, the interaction strategy should aim to solve problems and improve satisfaction. For example, if the behavior prediction module 240 predicts that the user's behavior type is to purchase a specific camera, and the current situation is the consideration stage (the interaction behavior data records that the user has viewed the camera's product page multiple times and has viewed the camera's review articles on social media platforms), and the emotion tendency index generated by the feature analysis module 230 indicates that the user's emotion tendency is positive, then the response generation module 250 can generate personalized response information that mimics the user's language style and is suitable for positive emotions, such as "The camera you are considering is currently on limited-time offer. Here is also a complete feature comparison table to help you quickly find the most suitable option," etc.

[0084] The response generation module 250 can also execute service recovery strategies when the response monitoring module 260 determines that a user has a negative experience. Service recovery strategies include, but are not limited to, reducing or stopping marketing activities for the user, marking the user as a high priority, providing compensatory coupons, and prompting live customer service to proactively contact the user.

[0085] The response monitoring module 260 is responsible for monitoring the user's interactive behavior after receiving the personalized response information generated by the response generation module 250, provided to the user by the data transmission module 210. The monitoring results may include performance indicators (such as positive response rate and action conversion rate), emotional change trends, and user interactive behavior (user's interactive behavior data within a certain period after receiving the personalized response information). User interactive behavior includes, but is not limited to, clicks, replies, browsing dwell time, emotional changes, and action conversion rate. For example, the response monitoring module 260 can continuously collect interactive behavior data (the user's interactive behavior data collected during this period is the user's interactive behavior) within a certain period of time after the user receives personalized response information. For example, it can track whether the user clicks on the links in the personalized response information through the information identification code in the personalized response information, calculate the time the user spends browsing the content recommended in the personalized response information (such as articles, videos, product information), whether the user replies to the content of the personalized response information (such as replying to the information, leaving a message, or publishing an article), or whether the user ignores the personalized response information, or whether the user follows the suggestions in the personalized response information (such as following customer service suggestions or adding items to the shopping cart), thereby generating monitoring results. The response monitoring module 260 can also analyze the user's emotional tendency when receiving personalized response information through the feature analysis module 230. For example, it can classify emotions based on the text of the user's reply, or detect emotional state from the acoustic characteristics of the voice call, in order to determine whether the personalized response information has triggered positive or negative psychological effects.

[0086] The response monitoring module 260 can also detect whether the user interaction behavior in the monitoring results includes a specific combination of behaviors, such as quickly closing messages, returning products, staying on the customer service page for a long time, or posting negative comments on social media. If so, it is determined that the user has had a negative experience.

[0087] The model adjustment module 270 is responsible for adjusting the parameters, such as weights or prediction rules, in the artificial intelligence model used by the behavior prediction module 240 based on the monitoring results generated by the response monitoring module 260 regarding the types of predicted behaviors. For example, the model adjustment module 270 can perform attribution analysis on the monitoring results to determine the key features affecting user interaction behavior and update the weights corresponding to those key features in the artificial intelligence model. For instance, if the attribution analysis reveals that the "emotional volatility" index has a high impact on shopping decisions, the model adjustment module 270 can increase the weights of features related to the "emotional volatility" index in the artificial intelligence model to improve the accuracy of future predictions. In this way, through the model adjustment module 270, the artificial intelligence model can learn and continuously optimize itself.

[0088] The system operation and operation method of the present invention will then be explained using an embodiment, and please refer to [reference needed]. Figure 3A The flowchart of the method for interaction and self-adjustment based on multi-platform data prediction behavior type proposed in this invention.

[0089] The data transmission module 210 of device 100 can acquire user interaction behavior data from multiple data sources (step 310). In this embodiment, it is assumed that... Figure 4 As shown, the data sources include social media platforms 411, e-commerce websites 412, and corporate websites 413. The data transmission module 210 can obtain the articles, comments, and likes posted by users on social media platforms 411, the product browsing history and shopping cart operation history on e-commerce websites 412, and the dialogue history with customer service on corporate websites 413. Furthermore, the data transmission module 210 can also obtain the user's past interaction data, such as the user's click rate on promotional notifications.

[0090] After the data transmission module 210 of device 100 obtains user interaction behavior data from multiple data sources (step 310), the data integration module 220 of device 100 can integrate all the interaction behavior data (step 320). In this embodiment, it is assumed that the data integration module 220 can standardize and de-identify the data format of the interaction behavior data from different data sources (social media platform 411, e-commerce website 412, and official website 413). Furthermore, the data integration module 220 can determine that the social media platform account and the e-commerce website member belong to the same user, and integrate the data obtained from the social media platform 411 and the e-commerce website 412 into complete cross-platform user data.

[0091] After the data integration module 220 of device 100 integrates the interactive behavior data, the feature analysis module 230 of device 100 can analyze the integrated interactive behavior data to generate user feature vectors and sentiment tendency indicators (step 330). Next, the behavior prediction module 240 of device 100 can input the user feature vectors and sentiment tendency indicators generated by the feature analysis module 230 as parameters into the artificial intelligence model, so that the artificial intelligence model generates the predicted behavior type of the user corresponding to the target platform (step 340). In this embodiment, it is assumed that the artificial intelligence model 275 can predict that the user has a "high probability of purchasing a specific camera" behavior type on all target platforms, or can predict the user's behavior type of purchasing a specific camera on e-commerce website 412, predict the user's behavior type of browsing articles or videos introducing a specific camera on social media platform 411, or predict the user's behavior type of asking customer service about the operation and maintenance of a specific product on the official website 413 of the company that manufactures the specific camera. Simultaneously, the behavior prediction module 240 can also... Figure 3B As shown in the flowchart, the user's current context is determined based on user intent and / or interaction behavior data (step 351).

[0092] After the behavior prediction module 240 of device 100 generates a predicted behavior type, the response generation module 250 of device 100 can generate personalized response information for each target platform to the user based on the predicted behavior type (step 350). In this embodiment, it is assumed that the response generation module 250 can generate personalized response information 422 that prompts the e-commerce website 412 (target platform) to recommend products related to a specific camera that the user has recently liked and / or commented on on the social media platform 411. Additionally, the response generation module 250 can also... Figure 3B As shown in the flowchart, natural language processing technology is used to analyze the user's public posts on social media in the interactive behavior data to generate a mirror communication style (step 353), such as "colloquial and with emojis". Based on the emotion tendency index generated by the feature analysis module 230 of device 100, the current situation and predicted behavior type generated by the behavior prediction module 240, and the historical interaction data obtained by the data transmission module 210, a corresponding interaction strategy is used to generate personalized response information that matches the generated mirror communication style (step 355). For example, personalized response information 421, "Hi! The camera you've been following recently is on sale today! Please refer to the following discount link. There are also extra lens bundles, limited quantity available." is generated and displayed on social media platform 411 and the company's official website 413 (target platform).

[0093] After the response generation module 250 of device 100 generates personalized response information for the user, the data transmission module 210 of device 100 can provide the personalized response information to the user. In this embodiment, such as Figure 3C As shown in the flowchart, the behavior prediction module 240 of device 100 can select the best interaction channel based on the predicted behavior type, the current situation, and the user's preferences (step 361). For example, it can select the App push notification with the highest click rate as the best interaction channel. Then, the data transmission module 210 can use the best interaction channel determined by the behavior prediction module 240 to provide personalized response information to the user (step 365). That is, the data transmission module 210 can use the mobile application of social media platform 411 and e-commerce website 412 to push personalized response information to the user.

[0094] After the data transmission module 210 of device 100 provides the user with the personalized response information generated by the response generation module 250 of device 100, the response monitoring module 260 of device 100 can monitor the user's interactive behavior on the target platform to generate monitoring results (step 370). In this embodiment, if the user receives a push notification from the mobile application of social media platform 411, opens the push notification, and clicks the discount link in the push notification to enter the shopping page of e-commerce website 412, the response monitoring module 260 can obtain the user's interactive behavior of clicking the link in the push notification based on the information identification code after the user clicks the link in the push notification, and can record the obtained user interactive behavior of clicking the link in the push notification into the monitoring results.

[0095] After the response monitoring module 260 of device 100 generates a monitoring result, the model adjustment module 270 of device 100 can adjust the parameters in the artificial intelligence model related to the type of predicted behavior generated based on the monitoring result (step 380). In this embodiment, it is assumed that the model adjustment module 270 can adjust the parameters in the artificial intelligence model related to the type of predicted behavior generated based on the monitoring result. Figure 3D As shown in the process, attribution analysis is performed on the monitoring results to determine the key features that affect user interaction behavior (step 381). For example, "short-term intent" is the most critical factor affecting user interaction behavior. Then, the model adjustment module 270 can update the weights in the artificial intelligence model corresponding to the key features of "short-term intent" (step 385) to increase the accuracy of future predictions.

[0096] Thus, through this invention, user interaction data across multiple platforms can be analyzed to generate personalized response information for users, and the artificial intelligence model can be adjusted based on the user's interaction behavior after receiving the personalized response information.

[0097] In the above embodiments, it is also possible to... Figure 3E As shown in the process, after the model adjustment module 270 of device 100 adjusts the parameters of the artificial intelligence model 275 regarding the type of predicted behavior based on the monitoring results generated by the response monitoring module 260 of device 100 (step 380), the response monitoring module 260 can detect whether the user's interaction behavior contains a specific combination of behaviors with negative experiences (step 391). If the user's interaction behavior records that the user publishes negative articles about products or companies on social media platform 411, or leaves negative reviews on the product page of e-commerce website 412, or stays on the customer service page of the company's official website 413 for a long time and enters negative messages, then the response monitoring module 260 can determine that the user has a negative experience, and the response generation module 250 of device 100 can execute a service recovery strategy (step 395), such as providing coupons or human customer service intervention.

[0098] In summary, the difference between this invention and existing technologies lies in its ability to analyze user interaction data across multiple platforms, predict possible user behavior types using an artificial intelligence model based on the analysis results, generate personalized response information based on the predicted behavior types, and transmit the response information to interact with the user. Furthermore, it can adjust the artificial intelligence model based on the user's interaction behavior. This approach addresses the problem in existing technologies where personalized information generated on online platforms is ineffective because it does not consider emotional biases or make real-time adjustments. Ultimately, this achieves the technical benefits of precise and personalized interaction with users and improved service efficiency.

[0099] Furthermore, the method of interactive and self-adjusting based on multi-platform data prediction behavior of the present invention can be implemented in hardware, software or a combination of hardware and software, or it can be implemented in a centralized manner in a computer system or in a decentralized manner with different components distributed among several interconnected computer systems.

[0100] 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 by those skilled in the art to the form and details of the implementation of this invention 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 scope defined in the appended claims.

Claims

1. A method for interactive and self-adjusting behavior based on multi-platform data prediction, applied to a device, the method comprising at least the following steps: Obtain user interaction data from multiple data sources; Integrate the aforementioned interactive behavior data; Analyze the interactive behavior data to generate user feature vectors and sentiment indexes; The user's feature vector and the sentiment index are input as parameters into the artificial intelligence model, so that the artificial intelligence model can generate the predicted behavior type of the user for each target platform. Based on the predicted behavior type, generate personalized response information for each target platform corresponding to the user; The monitoring process generates monitoring results by observing the user's interaction behavior after the personalized response information is provided to the user. and Adjust the parameters in the artificial intelligence model that generate the predicted behavior type based on the monitoring results.

2. The method for interaction and self-adjustment based on multi-platform data prediction behavior as described in claim 1, wherein before the step of monitoring the user's interaction behavior after the personalized response information is provided to the user, the method further includes the step of selecting the best interaction channel based on the predicted behavior type, the current context, and the user's user preferences, and using the best interaction channel to provide the personalized response information to the user.

3. The method for interactive and self-adjusting based on multi-platform data prediction behavior as described in claim 1, wherein the step of adjusting the parameters in the artificial intelligence model related to the generation of the predicted behavior type based on the monitoring results is to perform attribution analysis on the monitoring results to determine the key features affecting the user's interactive behavior, and update the weights in the artificial intelligence model corresponding to the key features.

4. The method for interactive and self-adjusting based on multi-platform data prediction behavior as described in claim 1, wherein the step of monitoring the user's interactive behavior after transmitting the personalized response information to generate the monitoring result further includes, when the monitoring result detects that the user's interactive behavior contains a specific combination of behaviors with negative experience, executing a service remediation strategy, the service remediation strategy including reducing or stopping marketing activities for the user, marking the user as a high priority level, providing compensatory coupons, and prompting a live customer service representative to proactively contact the user.

5. The method for interaction and self-adjustment based on multi-platform data prediction behavior as described in claim 1, wherein the step of generating the user's personalized response information based on the predicted behavior type and the user's historical interaction data further includes the steps of determining the user's current situation based on the user's intent determined from the context of the interaction behavior data and / or the user's behavior on the platform, analyzing the user's public posts on social media in the interaction behavior data using natural language processing technology to generate a mirror communication style, and generating personalized response information that conforms to the mirror communication style using corresponding interaction strategies based on the sentiment tendency index, the current situation, the predicted behavior type and the historical interaction data.

6. An apparatus for interactive and self-adjusting behavior based on multi-platform data prediction, the apparatus comprising at least: The data transmission module is used to acquire user interaction data from multiple data sources; The data integration module is used to integrate the interactive behavior data; The feature analysis module is used to analyze the interactive behavior data to generate user feature vectors and sentiment indexes; The behavior prediction module is used to input the user's feature vector and the sentiment tendency index as parameters into the artificial intelligence model, so that the artificial intelligence model can generate the predicted behavior type of the user corresponding to the target platform. The response generation module is used to generate personalized response information for each target platform and corresponding user based on the predicted behavior type. The response monitoring module is used to monitor the user's interaction behavior after the personalized response information is provided to the user in order to generate monitoring results; and The model adjustment module is used to adjust the parameters in the artificial intelligence model related to the type of predicted behavior based on the monitoring results.

7. The apparatus for interaction and self-adjustment based on multi-platform data prediction behavior as described in claim 6, wherein the behavior prediction module is further configured to select the optimal interaction channel based on the predicted behavior type, the current context, and the user preferences obtained by the data transmission module, and the data transmission module is further configured to use the optimal interaction channel to provide the personalized response information to the user.

8. The apparatus for interactive and self-adjusting based on multi-platform data prediction behavior as described in claim 6, wherein the model adjustment module performs attribution analysis on the monitoring results to determine the key features affecting the user's interactive behavior, and updates the weights corresponding to the key features in the artificial intelligence model.

9. The apparatus for interactive and self-adjusting based on multi-platform data prediction behavior as described in claim 6, wherein the response monitoring module is further configured to determine that the user has a negative experience when the monitoring results detect that the user's interactive behavior includes a specific combination of behaviors, and the response generation module is further configured to execute a service recovery strategy when the response monitoring module determines that the user has a negative experience, the service recovery strategy including reducing or stopping marketing activities for the user, marking the user as a high priority level, providing compensatory coupons, and prompting a live customer service representative to proactively contact the user.

10. The apparatus for interaction and self-adjustment based on multi-platform data prediction behavior as described in claim 6, wherein the response generation module is further configured to determine the user's current situation based on the user's intent determined by the behavior prediction module according to the context of the interaction behavior data and / or the user's behavior on the platform, and to analyze the user's public posts on social media in the interaction behavior data through natural language processing technology to generate a mirror communication style, and to generate personalized response information that conforms to the mirror communication style by using the corresponding interaction strategy based on the emotion tendency index, the current situation, the predicted behavior type and the historical interaction data.

Images