A human-computer interaction method based on intelligent data analysis

Abstract

This invention discloses a human-computer interaction method based on intelligent data analysis, belonging to the field of human-computer interaction technology. This invention collects user interaction actions on various interactive content, obtains multiple interaction records, constructs a user behavior dataset, adjusts the decay of interaction record data based on the regularity of the interaction content, and obtains the retention weight of interaction records based on the frequency of interaction actions, updating the user behavior dataset and the human-computer interaction model. The aim of this invention is to improve the efficiency of user interaction with the human-computer interaction model by updating the user behavior dataset and the human-computer interaction model, enabling the user behavior dataset to more accurately reflect user interaction preferences.

Description

Technical Field

[0001] This application relates to the field of human-computer interaction technology, and in particular to a human-computer interaction method based on intelligent data analysis. Background Technology

[0002] Human-computer interaction (HCI) refers to the interaction between humans and computer systems. Traditional HCI relied on users actively becoming accustomed to computer commands or graphical interfaces. However, with the development of big data technology and the improvement of computing power, modern HCI is increasingly inclined towards computers proactively adapting to users. By analyzing user behavior data, computers proactively understand user intentions, predict user needs, and optimize the interaction process. Analyzing user behavior data and providing feedback for optimization makes HCI more convenient and efficient.

[0003] In human-computer interaction in fields such as digital health, autonomous driving, and personalized recommendations, user behavior data analysis is a crucial step in ensuring the efficiency of human-computer interaction. This often requires real-time collection and analysis of user behavior data, updating the user behavior dataset, and preventing outdated data from affecting the accuracy of the system.

[0004] However, most existing human-computer interaction user behavior data analysis methods update the human-computer interaction model by periodically retraining the model in batches. Existing methods focus more on the time when the behavior data occurs, ignoring the regularity of long-term low-frequency data, and are easily affected by occasional data, leading to incorrect estimation of user behavior. Summary of the Invention

[0005] Therefore, it is necessary to provide a human-computer interaction method based on intelligent data analysis to address the aforementioned technical problems.

[0006] The following technical solution is adopted in this specification: This manual provides a human-computer interaction method based on intelligent data analysis, including: Each user selects an interactive item via touch control in the graphical user interface as an interaction behavior; a sequence of consecutive interaction behaviors within a preset time period is recorded as an interaction record; the time corresponding to the first interaction behavior in an interaction record is recorded as the occurrence time of that interaction record; multiple interaction records are obtained to construct a user behavior dataset. Based on the total number of interactions occurring for each interaction content in the user behavior dataset, obtain the frequency weight of each interaction content; Based on the similarity and temporal regularity of the interaction records containing each interaction content, obtain the regularity coefficient of each interaction content; Based on the occurrence time of each interaction record, the regularity of each interaction content in each interaction record, and the frequency weight of each interaction content in each interaction record, the retention weight of each interaction record is obtained. Based on the retention weights of all interaction records, the user behavior dataset is updated, and the human-computer interaction model is further updated.

[0007] Furthermore, obtaining the frequency weights of each interactive element specifically includes: When any two interactive elements exist in the same interactive record, the two interactive elements are considered related. The total number of interactive records in which the two related interactive elements exist simultaneously is recorded as the number of times the two interactive elements are related. Frequency weight of the k-th interaction item at time t The calculation method is as follows: In the formula: This represents the total number of interaction items related to the k-th interaction item at time t. This represents the number of times the m-th interaction item is related to the k-th interaction item at time t. This represents the number of interactions for the k-th interactive item at time t. This represents the number of interactions related to the m-th interactive item at time t, which is related to the k-th interactive item. This represents the total number of interactions for all items at time t.

[0008] Furthermore, the number of interactions specifically includes: The total number of times any interaction content occurs in all interaction records of the user behavior dataset at any given time is recorded as the interaction count of that interaction content at that time.

[0009] Furthermore, the acquisition of regularity coefficients for each interactive content specifically includes: The time interval between the occurrence of the f-th interaction record containing the k-th interaction content and the f+1-th interaction record containing the k-th interaction content is denoted as the interaction interval of the f-th interaction record containing the k-th interaction content. The regularity coefficient of the k-th interaction item at time t. The calculation method is as follows: In the formula: This represents the total number of interaction records containing the k-th interaction item at time t. This represents the total number of interaction intervals between the interaction records containing the k-th interaction item at time t. This represents the minimum matching distance between the f-th interaction record containing the k-th interaction content at time t and all other interaction records. This indicates a preset distance threshold. This represents the interaction interval of the f-th interaction record containing the k-th interaction item. This represents the average interaction interval across all interaction records containing the k-th interaction item. This represents an exponential function with the natural constant as its base.

[0010] Furthermore, the specific method for obtaining the matching distance is as follows: The distance between any two different interactive behaviors is recorded as 1. For any two interactive records, the DTW distance is calculated for the sequence of interactive behaviors that constitute these two interactive records. The obtained DTW distance result is recorded as the matching distance between these two interactive records.

[0011] Furthermore, obtaining the retention weight of each interaction record specifically includes: The interval between the occurrence time of the r-th interaction record and the t-th time is denoted as the decay time of the r-th interaction record; The retention weight of the r-th interaction record at time t. The calculation method is as follows: In the formula: This represents the average frequency weight of all interactive behaviors corresponding to the interactive content in the r-th interaction record at time t. This represents the suppression coefficient of the r-th interaction record at time t. This represents the preset attenuation constant. This represents the decay time of the r-th interaction record at time t.

[0012] Furthermore, the specific method for obtaining the inhibition coefficient is as follows: The suppression coefficient of the r-th interaction record at time t. The calculation method is as follows: In the formula: This represents the total number of interaction items in the r-th interaction record at time t. Let represent the regularity coefficient of the k-th interaction content in the r-th interaction record at time t. This represents the average interaction interval of all interaction records containing the k-th interaction content in the r-th interaction record at time t. This represents the maximum decay time of the interaction record containing all interactive content at time t.

[0013] Furthermore, updating the user behavior dataset specifically includes: A preset retention threshold and an update threshold are set. When the retention weight of any interaction record at time t is less than the retention threshold, the interaction record is recorded as a failed record. When the proportion of the total number of failed records to the total number of all interaction records at time t exceeds the update threshold, all failed records are removed from the user behavior dataset in ascending order of retention weight until the proportion of the total number of failed records to the total number of all interaction records at time t is less than or equal to the update threshold, or the total number of interaction records in the user behavior dataset is equal to the minimum collection amount. Then the update of the user behavior dataset is completed, and the updated user behavior dataset is obtained.

[0014] Furthermore, the updating of the human-computer interaction model specifically includes: Based on the priority weight of the k-th interactive content after the update at time t, all interactive content is sorted in descending order of priority weight, and the human-computer interaction graphical interface is updated to facilitate users to select interactive content more efficiently.

[0015] Furthermore, the specific method for obtaining the priority weight is as follows: The priority weight of the k-th interaction item after the update at time t. The calculation method is as follows: In the formula: This represents the total number of all interaction records after the update at time t. This represents the retention weight of the j-th interaction record at time t. This represents the existence status of the k-th interaction item in the j-th interaction record after the update at time t. It specifies that when the j-th interaction record contains the k-th interaction item, then... Conversely, when the j-th interaction record contains no k-th interaction content after the t-th time update, .

[0016] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: In the human-computer interaction method based on intelligent data analysis provided in this specification, the retention weight of interaction record data is evaluated based on the similarity and regularity of user interaction behaviors and the occurrence time of interaction records, thereby determining the update rules of the dataset; this avoids ignoring low-frequency regular data and also avoids the excessive influence of occasional data, so that the retained interaction record data can more accurately reflect user interaction preferences. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0018] Figure 1 This document provides a flowchart illustrating a human-computer interaction method based on intelligent data analysis. Figure 2 This is a diagram illustrating the priority of the interactive pages provided in this manual. Figure 3 This is a schematic diagram of the application process in the online customer service scenario provided in this manual. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0020] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0021] Example 1 Figure 1 This is a flowchart illustrating a human-computer interaction method based on intelligent data analysis as described in this specification, which specifically includes the following steps: S101: Collect user interaction behavior data and construct a user behavior dataset.

[0022] The purpose of this embodiment is to use user interaction behavior data to update the user interaction model based on intelligent data analysis. Therefore, it is first necessary to collect user interaction behavior data and construct a user behavior dataset.

[0023] The user interaction behavior data mentioned in this embodiment can be human-computer interaction behavior data implemented based on interaction methods such as voice control or touch control. For ease of explanation, the following description will only focus on touch control as the execution subject.

[0024] The preset interruption time is 10 minutes, and the preset minimum collection quantity is 100 records. These values ​​are for illustrative purposes only and can be adjusted according to the specific implementation. Each user's selection of an interactive content via touch control in the graphical user interface is recorded as an interaction. When the interval between a user's current interaction and its preceding interaction exceeds the interruption time, the interval is recorded as a no-interaction state. The sequence of all user interactions between two adjacent no-interaction states is recorded as an interaction record. The time corresponding to the first interaction in an interaction record is recorded as the occurrence time of that interaction record.

[0025] Specifically, every 1 second, a time point is selected, and at least the minimum amount of interaction records are collected. Taking the t-th time point as an example, the user behavior dataset for the t-th time point is constructed using all interaction records before the t-th time point.

[0026] S102: Obtain the frequency weight of each interactive content at each time point.

[0027] It should be noted that the more interactive content a user generates, the higher the user's tendency to use that interactive content, and the higher the priority of that interactive content in the human-computer interaction model. In addition, some interactive content itself has a low number of interactive behaviors, but exists in the same interaction record as other interactive content with a high number of interactive behaviors. In this case, the interactive content is influenced by the other interactive content with a high number of interactive behaviors, and its priority will also increase.

[0028] In the user behavior dataset at any given time, the total number of interactions for any given interaction is recorded as the interaction count for that interaction at that time. If any two interactions exist in the same interaction record, they are considered related, and the total number of interaction records where these two related interactions coexist is recorded as the correlation count between them. This step obtains the frequency weight of each interaction based on its interaction count and correlation count with other interactions in the current user behavior dataset.

[0029] Specifically, taking the frequency weight of the k-th interaction at time t as an example, the frequency weight of the k-th interaction at time t is obtained based on the correlation counts of the k-th interaction with other interaction counts at time t, and the interaction count of the interaction itself. The calculation method is as follows: In the formula: This represents the total number of interaction items related to the k-th interaction item at time t. This represents the number of times the m-th interaction item is related to the k-th interaction item at time t. This represents the number of interactions for the k-th interactive item at time t. This represents the number of interactions related to the m-th interactive item at time t, which is related to the k-th interactive item. This represents the total number of interactions for all items at time t.

[0030] It should be noted that This represents the proportion of the number of times the k-th interactive content is related to the m-th interactive content at time t in the total number of interactions of the k-th interactive content. It reflects the degree of correlation between the k-th and m-th interactive content. The larger this value is, the greater the degree of correlation between the k-th and m-th interactive content. This represents the percentage of interactions with the m-th interactive content related to the k-th interactive content at time t, reflecting the level of interaction with the m-th interactive content related to the k-th interactive content at time t. It also indicates the user's preference for this interactive content at time t; a larger value indicates a stronger user preference. The overall relevance and interaction frequency of all interactive content related to the k-th interactive content at time t are used as weights to influence the interaction frequency of the k-th interactive content itself. Gain the frequency weight of the k-th interaction item at time t.

[0031] It should be noted that the frequency weights of various interactive content obtained from the user behavior dataset at each time point can represent the user's preference for each interactive content at each time point. The higher the frequency weight of the interactive content, the higher the user's preference. Therefore, by prioritizing the display of interactive content with high user preference on the human-computer interaction interface based on the preference reflected by the frequency weight, a more convenient interaction process can be provided for the user.

[0032] S103: Obtain the regularity coefficients of various interactive contents at each time point.

[0033] Frequency weighting reflects user preference based on the number of interactions. However, if frequency weighting is solely based on the number of interactions, it fails to reflect the impact of the regularity of user interaction behavior on user preference. When evaluating user preferences for interactions with low frequency but certain regularity, the resulting user preference level is too low. Therefore, this step obtains the regularity coefficient of each interaction based on the similarity and temporal regularity of the interaction records. The more similar the different interaction records containing each interaction, and the closer the time interval between the occurrences of different interaction records containing each interaction, the stronger the regularity and the larger the regularity coefficient.

[0034] Specifically, this step uses the DTW distance to represent the similarity between different interaction records. The distance between any two different interaction behaviors is recorded as 1. For any two interaction records, the DTW distance is calculated for the sequence of interaction behaviors that constitute these two interaction records. The obtained DTW distance result is recorded as the matching distance between these two interaction records. The method for obtaining the DTW distance is existing technology and will not be described in detail in this embodiment. A preset distance threshold is used. When the matching distance between two interaction records is less than the distance threshold, the two interaction records are considered to have a high similarity. The value of the distance threshold here is only an example and can be adjusted according to the specific situation.

[0035] Furthermore, this step uses the difference in the occurrence time interval between adjacent interaction records containing the interaction content to reflect the temporal regularity of the interaction behavior. Taking the f-th interaction record containing the k-th interaction content as an example, the time interval between the occurrence time of the f-th interaction record containing the k-th interaction content and the (f+1)-th interaction record containing the k-th interaction content is recorded as the interaction interval of the f-th interaction record containing the k-th interaction content. As shown in Table 1, a comparison table of the temporal regularity of interaction content is presented. In the table, the interaction intervals of interaction content 1 are similar, while the interaction intervals of interaction content 2 differ significantly. Therefore, the temporal regularity of interaction content 1 is stronger.

[0036] Table 1. Comparison of Temporal Regularities in Interactive Content Furthermore, based on the similarity between different interaction records containing the k-th interaction content at time t and the temporal regularity of the interaction behavior containing the k-th interaction content, the regularity coefficient of the k-th interaction content at time t is obtained; the regularity coefficient of the k-th interaction content at time t... The calculation method is as follows: In the formula: This represents the total number of interaction records containing the k-th interaction item at time t. This represents the total number of interaction intervals between the interaction records containing the k-th interaction item at time t. This represents the minimum matching distance between the f-th interaction record containing the k-th interaction content at time t and all other interaction records. This indicates a preset distance threshold. This represents the interaction interval of the f-th interaction record containing the k-th interaction item. This represents the average interaction interval across all interaction records containing the k-th interaction item. This represents an exponential function with the natural constant as its base, which is used here. The model implements inverse proportional normalization.

[0037] In particular, when When the time interval between the occurrence time of the interaction record containing the k-th interaction content at time t and time t is recorded as the interaction interval.

[0038] It should be noted that This value represents the similarity between the f-th interaction record containing the k-th interaction content at time t and other interaction records containing the k-th interaction content. The smaller the value, the more similar the other interaction records containing the k-th interaction content are to the f-th interaction record. It represents the overall similarity level between each interaction record containing the k-th interaction content and other interaction records at time t. The smaller the value, the greater the overall similarity, and thus the greater the regularity coefficient. It represents the fluctuation level of the interaction interval. The smaller the value, the smaller the fluctuation of the interaction interval of each interaction record containing the k-th interaction content at time t, and thus the larger the regularity coefficient.

[0039] It should be noted that the regularity coefficients of various interaction contents at each time point reflect the regularity of the interaction records in the user behavior dataset at each time point. The regularity coefficients, through the regularity shown by various interaction contents, can more accurately analyze user behavior habits and preferences, avoid ignoring low-frequency regular data, and also avoid the excessive influence of occasional data.

[0040] S104: Obtain the retention weight of each interaction record.

[0041] User preferences for various interactive content may change over time. Therefore, earlier interaction records show greater attenuation in their relevance to analyzing user behavior and preferences, while later records are more likely to reflect the user's latest interaction habits. Existing technologies often use time decay functions to quantify the decay of historical data's reference value. These functions only consider the impact of time. However, when interaction records contain highly regular patterns, relying solely on time for decay may lead to an overestimation of the decay of interaction records that best reflect user habits. Therefore, this step optimizes the existing time decay function based on the regularity of interaction content in each record, obtaining the retention weight for each interaction record at each time point.

[0042] Specifically, if time t is the current time, the interval between the occurrence time of the r-th interaction record and time t is recorded as the decay time of the r-th interaction record; based on the decay time of each interaction record and the regularity of the interaction content in the interaction record, the inhibition coefficient of each interaction record at each time is obtained; the inhibition coefficient of the r-th interaction record at time t. The calculation method is as follows: In the formula: This represents the total number of interaction items in the r-th interaction record at time t. Let represent the regularity coefficient of the k-th interaction content in the r-th interaction record at time t. This represents the average interaction interval of all interaction records containing the k-th interaction content in the r-th interaction record at time t. This represents the maximum decay time of the interaction record containing all interactive content at time t.

[0043] It should be noted that It shows the regularity of the interactive content. The stronger the regularity, the more likely the user has a regular preference for the interactive content, and the more necessary it is to suppress the decay caused by time. It represents the size of the overall interaction interval of the interactive content. The larger the interaction interval, the more necessary it is to suppress the decay caused by time.

[0044] Furthermore, the time decay function is optimized by combining the suppression coefficient of each interaction record to obtain the retention weight of each interaction record at each time step; the retention weight of the r-th interaction record at time step t is... The calculation method is as follows: In the formula: This represents the average frequency weight of all interactive behaviors corresponding to the interactive content in the r-th interaction record at time t. This represents the suppression coefficient of the r-th interaction record at time t. This represents the preset attenuation constant. Here, we take a preset value of 0.05 as an example. The value can be adjusted according to specific circumstances. This represents the decay time of the r-th interaction record at time t.

[0045] It should be noted that the exponential decay function of the time decay function in the prior art can be expressed as: The exponential decay function is existing technology and will not be elaborated upon in this embodiment. This step optimizes the exponential decay function based on the existing technology, combining the regularity coefficient of the interactive content and the interaction interval. In the existing technology, the initial state of decay is a fixed value, while this step uses... This initial state serves as the basis for attenuation, making the results more consistent with the user's dynamic preferences for different interactive content. In existing technologies, the attenuation constant is a fixed value, while this step uses... As a coefficient adjustment decay constant, the decay of highly regular interaction records is suppressed. Similarly, the retention weight of each interaction record at each time point is obtained.

[0046] It should be noted that the retention weights obtained in this step based on the exponential decay function of existing technology take into account the decay caused by time, and obtain the retention weights of each interaction record at each time point. The retention weights represent the reference value of each interaction record for user interaction habits at different times. The acquisition of retention weights combines the regularity of interaction records, which can better retain interaction records with regularity, so that the retained interaction records can more accurately reflect user interaction preferences.

[0047] S105: Update the human-computer interaction model based on the retention weights of all interaction records at each time point.

[0048] By constructing a user behavior dataset, evaluating the retention weight of each interaction record in the dataset, and updating the user behavior dataset based on the overall retention weight level of the interaction records, the user behavior dataset can reflect the user's real-time preference state. Based on the user's real-time preference state, the human-computer interaction model is updated, thereby improving the efficiency of human-computer interaction.

[0049] Specifically, the preset retention threshold is 0.25 and the update threshold is 0.1. These values ​​are only examples and can be adjusted according to specific circumstances. Taking time t as an example, when the retention weight of any interaction record at time t is less than the retention threshold, the interaction record is recorded as an invalid record. When the total number of invalid records accounts for more than the proportion of the total number of all interaction records at time t, the current user behavior dataset is considered to need to be updated.

[0050] Furthermore, if the current user behavior dataset needs to be updated, all invalid records are removed from the user behavior dataset in ascending order of retention weight until the total number of invalid records is less than or equal to the percentage of the total number of all interaction records at time t, or the total number of interaction records in the user behavior dataset is equal to the minimum collection amount. Then the update of the user behavior dataset is completed, and the updated user behavior dataset is obtained.

[0051] This completes the update of the user behavior dataset in the human-computer interaction model.

[0052] Optionally, by obtaining the retention weight of each interaction record at each time point, the reference value of each interaction record for analyzing user interaction preferences is quantified. The higher the retention weight of the interaction record containing the interaction content, the stronger the user's preference for that interaction content. The user preference degree of each interaction content is reflected by the interaction records containing each interaction content in the updated user behavior dataset, and is recorded as the priority weight of each interaction content. The visualization priority of the graphical interface of the human-computer interaction model is updated based on the priority weight of each interaction content.

[0053] Specifically, based on the retention weights of all interaction records in the updated user behavior dataset and the presence of each interaction item in each interaction record, the priority weight of each interaction item is obtained; if the user behavior dataset is updated at time t, then the priority weight of the k-th interaction item after the update at time t is... The calculation method is as follows: In the formula: This represents the total number of all interaction records after the update at time t. This represents the retention weight of the j-th interaction record at time t. This represents the existence status of the k-th interaction item in the j-th interaction record after the update at time t. It specifies that when the j-th interaction record contains the k-th interaction item, then... Conversely, when the j-th interaction record contains no k-th interaction content after the t-th time update, Similarly, obtain the priority weights of all item interactions after the update at time t.

[0054] It should be noted that the priority weights are based on the distribution of various interactive content in the updated user behavior dataset, which more accurately reflects the user's preference for each interactive content. The higher the priority weight, the stronger the user's preference for that interactive content.

[0055] Furthermore, based on the priority weight of the k-th interactive content after the update at time t, all interactive content is sorted in descending order of priority weight, and the human-computer interaction graphical interface is updated to facilitate users in selecting interactive content more efficiently; such as Figure 2 The diagram shows the priority of interactive pages. As an example, the content in interactive page 1 has a higher priority weight than that in interactive page 2. The priority weight is used as the basis for page division. The specific arrangement of interactive content can be based on the priority weight in descending order or other methods. There are no specific restrictions here.

[0056] This concludes the embodiment.

[0057] Example 2 This embodiment provides a specific application of a human-computer interaction method based on intelligent data analysis in an online customer service scenario, such as... Figure 3 The diagram shown illustrates the application process in the online customer service scenario provided in this manual.

[0058] Online customer service systems are an important application area of ​​human-computer interaction, where users solve problems by interacting with customer service robots or human customer service representatives. This embodiment aims to analyze the historical interaction behavior of users with the customer service system, dynamically evaluate the reference value of each interaction record, thereby optimizing the response strategy of the customer service robot and the display order of frequently asked questions (FAQs), and improving user consultation efficiency and satisfaction.

[0059] S201: Collect user interaction data with the customer service system and build a user behavior dataset.

[0060] In this embodiment, each interaction between the user and the online customer service system is considered a human-computer interaction process. The specific definition is as follows:

[0061] Each user's message (including text, emoticons, images, etc.) sent in the customer service chat window or a click on a preset question option is recorded as an interaction. The preset interruption time is 5 minutes. That is, when the time interval between the user's current message and the previous message exceeds 5 minutes, it is considered a "no-interaction state", indicating the interruption of a consultation session or the user's temporary departure. The sequence of all user interactions between two adjacent no-interaction states is recorded as an interaction record, which represents a complete customer service consultation session. The time when the first message in an interaction record is sent is recorded as the occurrence time of that interaction record.

[0062] The system continuously collects user interaction behavior to build a user behavior dataset. Each interaction record in the dataset includes: the time of occurrence, the sequence of interaction behavior (e.g., ["click on shipping fee issue", "enter order number", "inquire about delivery time"]), and the interaction content corresponding to each message (i.e. the customer service issue category that the message points to, such as "logistics inquiry", "return and exchange policy", "product usage guide", etc.).

[0063] S202: Obtain the frequency weight of each customer service question category at each time point.

[0064] It should be noted that the more frequently a question category is consulted by users, the higher the user's attention to that type of question, and it should have a higher priority in the FAQ display of the customer service robot. In addition, some question categories themselves do not have a high number of consultations, but they often appear in the same conversation as question categories with high consultation frequency (for example, a user asks "how to place an order" first, and then asks "how to use coupons"), then the former is affected by the latter, and its priority should also be increased.

[0065] In the user behavior dataset at any given time, the total number of times any customer service question category appears is recorded as the number of interactions for that question category at that time. When any two customer service question categories exist in the same interaction record, the two question categories are considered to be related, and the total number of interaction records in which the two question categories exist simultaneously is recorded as the number of related interactions between the two question categories.

[0066] This step obtains the frequency weight of each customer service question category based on the number of interactions and correlations with other question categories in the user behavior dataset at the current moment. The frequency weight reflects the user's overall preference for different consultation questions, providing a basis for optimizing the FAQ display order in the future.

[0067] S203: Obtain the regularity coefficients of various customer service question categories at each time point.

[0068] Frequency weighting only reflects the total number of inquiries about a problem, but it cannot reflect the regularity of user inquiry behavior. For example, some users may only inquire about "logistics progress" on Mondays. Although the total number of inquiries is not high, it has a clear periodic pattern and should be given higher priority in the corresponding time period. Therefore, this step obtains the regularity coefficient of each problem category based on the similarity and time regularity between different conversations of each customer service problem category.

[0069] Specifically, for any two interaction records, the DTW distance is used to measure their similarity in the sequence of interaction behaviors; the distance between any two different interaction behaviors is recorded as 1, and the DTW distance of the behavior sequences of the two session records is calculated as the matching distance between them; the smaller the matching distance, the more similar the consultation paths of the two sessions are.

[0070] Furthermore, for the same customer service question category, the time interval between two adjacent sessions is calculated based on the occurrence time of the multiple interaction records in which it is located, and is recorded as the interaction interval; the more stable the interaction interval, the stronger the time regularity of the consultation behavior for this question category.

[0071] Therefore, the minimum matching distance reflects the similarity between different sessions of the same question category. Combined with the stability of the interaction interval, the regularity coefficient of each question category is obtained. The larger the regularity coefficient, the stronger the regularity of the consultation behavior of that question category, and it should be protected in the subsequent data decay.

[0072] S204: Obtain the retention weight of each interaction record.

[0073] Users' consultation preferences may change over time. For example, they may focus on consulting about "new product launches" for a period of time, and then shift their focus after a while. Therefore, the earlier the conversation record, the less valuable it is for analyzing current user preferences. Traditional time decay functions only consider the time factor, but if a conversation record contains a highly regular question category, such as a user's regular bill inquiry or logistics inquiry, it should not decay too quickly.

[0074] This step first calculates the decay time of each interaction record, that is, the interval between the occurrence time of the session record and the current time. Based on the regularity coefficients of each question category in the session record and their average interaction interval, the inhibition coefficient of the session record is obtained. The larger the inhibition coefficient, the more likely the session record should be decayed due to containing highly regular consultation questions.

[0075] Finally, the traditional exponential decay function is optimized by combining the inhibition coefficient to obtain the retention weight of each interaction record at each time point. The retention weight integrates the consultation content preference of the conversation itself, the regularity of the conversation, and the time decay factor, and quantifies the reference value of each historical conversation record for the analysis of current user preferences.

[0076] S205: Update the customer service robot's response model and FAQ display strategy based on the retention weight of all interaction records at each time point.

[0077] By constructing a user behavior dataset, the retention weight of each customer service conversation record is evaluated. When the proportion of low-weight conversation records is too high, a dataset update mechanism is triggered to remove outdated and occasional conversation data, retaining valid data that reflects the user's current consultation preferences.

[0078] Based on the updated user behavior dataset, priority weights are calculated for each customer service question category. The priority weight calculation combines the retention weight of each valid session record with the occurrence of that question category in the session; the higher the priority weight, the stronger the user's current consultation preference for that question category.

[0079] Ultimately, the system dynamically adjusts the online customer service robot's response strategy based on the priority weight of each question category, specifically as follows: FAQ Smart Sorting: In the FAQ list in the customer service chat window, questions with higher priority are displayed first, allowing users to quickly click to consult without typing. Robot knowledge base update: Answers to question categories with high priority weights will be loaded first into the hot spots of the robot knowledge base, shortening response time; Session routing optimization: For complex issues with high priority weight that the robot cannot solve, the system can prioritize the allocation of human customer service resources.

[0080] Through the above steps, this embodiment realizes the dynamic perception and adaptive optimization of user consultation preferences in the online customer service system, avoids the system ignoring low-frequency regular consultations, and filters out the interference of occasional one-off consultations on the model, thereby improving the efficiency of customer service interaction and user experience.

[0081] This concludes the example.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

[0083] based on Figure 1 The human-computer interaction method shown here is based on intelligent data analysis. It optimizes the update rules of user interaction data, avoids ignoring low-frequency regular data, and avoids the excessive influence of occasional data. This allows the interaction record data retained in the user behavior dataset to more accurately reflect user interaction preferences.

[0084] When applying the human-computer interaction method based on intelligent data analysis provided in this manual, it is not necessary to... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this manual does not impose any restrictions on it.

[0085] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A human-computer interaction method based on intelligent data analysis, characterized in that, include: Each interaction is recorded as a user selecting an interactive item via touch control in the graphical user interface. A sequence of consecutive interactions within a preset time period is recorded as an interaction record. The time corresponding to the first interaction in an interaction record is recorded as the occurrence time of that interaction record. Obtain multiple interaction records and construct a user behavior dataset; Based on the total number of interactions occurring for each interaction content in the user behavior dataset, obtain the frequency weight of each interaction content; Based on the similarity and temporal regularity of the interaction records containing each interaction content, obtain the regularity coefficient of each interaction content; Based on the occurrence time of each interaction record, the regularity of each interaction content in each interaction record, and the frequency weight of each interaction content in each interaction record, the retention weight of each interaction record is obtained. Based on the retention weights of all interaction records, the user behavior dataset is updated, and the human-computer interaction model is further updated.

2. The human-computer interaction method based on intelligent data analysis as described in claim 1, characterized in that, The acquisition of frequency weights for each interactive element specifically includes: When any two interactive elements exist in the same interactive record, the two interactive elements are considered related. The total number of interactive records in which the two related interactive elements exist simultaneously is recorded as the number of times the two interactive elements are related. Frequency weight of the k-th interaction item at time t The calculation method is as follows: In the formula: This represents the total number of interaction items related to the k-th interaction item at time t. This represents the number of times the m-th interaction item is related to the k-th interaction item at time t. This represents the number of interactions for the k-th interactive item at time t. This represents the number of interactions related to the m-th interactive item at time t, which is related to the k-th interactive item. This represents the total number of interactions for all items at time t.

3. The human-computer interaction method based on intelligent data analysis as described in claim 2, characterized in that, The number of interactions specifically includes: The total number of times any interaction content occurs in all interaction records of the user behavior dataset at any given time is recorded as the interaction count of that interaction content at that time.

4. The human-computer interaction method based on intelligent data analysis as described in claim 1, characterized in that, The acquisition of regularity coefficients for each interactive content specifically includes: The time interval between the occurrence of the f-th interaction record containing the k-th interaction content and the f+1-th interaction record containing the k-th interaction content is denoted as the interaction interval of the f-th interaction record containing the k-th interaction content. The regularity coefficient of the k-th interaction item at time t. The calculation method is as follows: In the formula: This represents the total number of interaction records containing the k-th interaction item at time t. This represents the total number of interaction intervals between the interaction records containing the k-th interaction item at time t. This represents the minimum matching distance between the f-th interaction record containing the k-th interaction content at time t and all other interaction records. This indicates a preset distance threshold. This represents the interaction interval of the f-th interaction record containing the k-th interaction item. This represents the average interaction interval across all interaction records containing the k-th interaction item. This represents an exponential function with the natural constant as its base.

5. The human-computer interaction method based on intelligent data analysis as described in claim 4, characterized in that, The specific method for obtaining the matching distance is as follows: The distance between any two different interactive behaviors is recorded as 1. For any two interactive records, the DTW distance is calculated for the sequence of interactive behaviors that constitute these two interactive records. The obtained DTW distance result is recorded as the matching distance between these two interactive records.

6. The human-computer interaction method based on intelligent data analysis as described in claim 4, characterized in that, The process of obtaining the retention weight of each interaction record specifically includes: The interval between the occurrence time of the r-th interaction record and the t-th time is denoted as the decay time of the r-th interaction record; The retention weight of the r-th interaction record at time t. The calculation method is as follows: In the formula: This represents the average frequency weight of all interactive behaviors corresponding to the interactive content in the r-th interaction record at time t. This represents the suppression coefficient of the r-th interaction record at time t. This represents the preset attenuation constant. This represents the decay time of the r-th interaction record at time t.

7. The human-computer interaction method based on intelligent data analysis as described in claim 6, characterized in that, The specific method for obtaining the inhibition coefficient is as follows: The suppression coefficient of the r-th interaction record at time t. The calculation method is as follows: In the formula: This represents the total number of interaction items in the r-th interaction record at time t. Let represent the regularity coefficient of the k-th interaction content in the r-th interaction record at time t. This represents the average interaction interval of all interaction records containing the k-th interaction content in the r-th interaction record at time t. This represents the maximum decay time of the interaction record containing all interactive content at time t.

8. The human-computer interaction method based on intelligent data analysis as described in claim 6, characterized in that, The update of the user behavior dataset specifically includes: A preset retention threshold and an update threshold are set. When the retention weight of any interaction record at time t is less than the retention threshold, the interaction record is recorded as a failed record. When the proportion of the total number of failed records to the total number of all interaction records at time t exceeds the update threshold, all failed records are removed from the user behavior dataset in ascending order of retention weight until the proportion of the total number of failed records to the total number of all interaction records at time t is less than or equal to the update threshold, or the total number of interaction records in the user behavior dataset is equal to the minimum collection amount. Then the update of the user behavior dataset is completed, and the updated user behavior dataset is obtained.

9. The human-computer interaction method based on intelligent data analysis as described in claim 8, characterized in that, The updated human-computer interaction model specifically includes: Based on the priority weight of the k-th interactive content after the update at time t, all interactive content is sorted in descending order of priority weight, and the human-computer interaction graphical interface is updated to facilitate users to select interactive content more efficiently.

10. The human-computer interaction method based on intelligent data analysis as described in claim 9, characterized in that, The specific method for obtaining the priority weight is as follows: The priority weight of the k-th interaction item after the update at time t. The calculation method is as follows: In the formula: This represents the total number of all interaction records after the update at time t. This represents the retention weight of the j-th interaction record at time t. This represents the existence status of the k-th interaction item in the j-th interaction record after the update at time t. It specifies that when the j-th interaction record contains the k-th interaction item, then... Conversely, when the j-th interaction record contains no k-th interaction content after the t-th time update, .

Images