An adaptive push system for hang-up messages based on remote communication

Abstract

This application relates to the field of communication message push technology, specifically to an adaptive push system for hang-up messages based on remote communication. The system includes: a communication data processing module for text recognition and word segmentation of information during each remote communication process; obtaining the word polarity of each segmented word in each occurrence; a communication feature extraction module for determining the information tendency and word tendency change rate of each segmented word during each remote communication process; obtaining the word temporal interference rate of each segmented word during each remote communication process; determining the trend change of each segmented word during the current remote communication process; and a hang-up message push module for obtaining the weight coefficient of each segmented word during the current remote communication process; correcting the TF value in the TF-IDF algorithm to obtain the TF-IDF value of each segmented word during the current remote communication process; filtering each segmented word using the TF-IDF value and matching it with recommended information to push the hang-up message. This application improves the accuracy of hang-up message push.

Description

Technical Field

[0001] This application relates to the field of communication message push technology, specifically to an adaptive push system for hang-up messages based on remote communication. Background Technology

[0002] "Away from activity" messages typically refer to messages automatically sent to users when they leave or become inactive on some online services or platforms. These messages aim to improve user experience, increase user engagement, and enhance service efficiency.

[0003] Existing push notifications for disconnected messages often suffer from insufficient understanding of user needs and an inability to accurately grasp users' true requirements. The push algorithms rely too heavily on historical data and fail to consider the user's current context, resulting in push notifications that may be irrelevant to the user's current interests and needs. Furthermore, the TF-IDF algorithm relies solely on word frequency when calculating the importance of word segments, leading to low reliability in extracting information of interest to users in remote communication, and consequently, low accuracy in push notifications for disconnected messages. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides an adaptive push system for hang-up messages based on remote communication, thereby resolving the existing issues.

[0005] The adaptive push system for hang-up messages based on remote communication proposed in this application adopts the following technical solution:

[0006] One embodiment of this application provides an adaptive push system for hang-up messages based on remote communication, the system comprising:

[0007] The communication data processing module is used to perform text recognition and word segmentation on the information during each remote communication process; and to obtain the word polarity of each word segmentation based on the sentiment tendency of each word segmentation.

[0008] The communication feature extraction module is used to determine the information tendency of each word in each remote communication process by combining the word polarity of all occurrences of each word in each remote communication process; analyze the difference of the information tendency of each word in each remote communication process relative to the first remote communication, and determine the word tendency change rate of each word in each remote communication process;

[0009] By weighting the word segmentation tendency change rate according to the importance in time, the word segmentation temporal interference rate of each word in each remote communication process is obtained; the change trend of the word segmentation temporal interference rate of each word in all previous remote communication processes is analyzed to determine the trend change amount of each word in the current remote communication process.

[0010] The message push module for hanging up (or ignoring communication) obtains the weight coefficient of each word segment in the current remote communication process based on the distribution of information tendency in the historical remote communication process and the trend change amount. It then corrects the TF value in the TF-IDF algorithm using the weight coefficient and uses the corrected TF-IDF algorithm to obtain the TF-IDF value of each word segment in the current remote communication process.

[0011] Each word segment is filtered using the TF-IDF value and matched with the recommended information to push out the AFK message.

[0012] In one embodiment, obtaining the word polarity of each occurrence of each segmented word includes:

[0013] The semantic analysis algorithm is used to obtain the sentiment classification results of each word segmentation at each occurrence. The word polarity of the word segmentation belonging to positive words is set to 1, the word polarity of the word segmentation belonging to negative words is set to -1, and the word polarity of the word segmentation belonging to neither positive nor negative words is set to 0.

[0014] In one embodiment, determining the information tendency includes:

[0015] For each remote communication, the sum of word polarity for all occurrences of each word segment is calculated. The information tendency is positively correlated with the sum and the number of occurrences of each word segment.

[0016] In one embodiment, the word segmentation tendency change rate is positively correlated with the difference between each word segmentation and negatively correlated with the information tendency of each word segmentation during the first remote communication process.

[0017] In one embodiment, determining the word segmentation temporal interference rate includes:

[0018] Calculate the sum of the sequence numbers of the current and all previous remote communication processes, calculate the ratio of the sequence number of each remote communication process to the sum, and the word segmentation temporal interference rate is the product of the word segmentation tendency change rate and the ratio.

[0019] In one embodiment, determining the amount of trend change includes:

[0020] Obtain the slope of the fitted straight line of the temporal interference rate of any segmentation in the current and all previous remote communication processes;

[0021] Calculate the sign result of the sum of the temporal interference rates of any segment in the current and all previous remote communication processes for any segment, and the trend change of any segment in the current remote communication process is the product of the slope and the sign result.

[0022] In one embodiment, determining the weighting coefficients includes:

[0023] The normalized result of the mean of the information tendency of each word in the current remote communication process in all previous remote communication processes is calculated, and the weight coefficient is the product of the normalized result and the normalized result of the trend change.

[0024] In one embodiment, the TF value in the TF-IDF algorithm is corrected using the weighting coefficients, and the calculation method is as follows:

[0025] In the formula, tf(t) is the TF value after the t-th word segmentation correction in the current remote communication process, and W t Let f(t) be the weight coefficient of the t-th word in the current remote communication process, f(t) be the frequency of the t-th word in the current remote communication process, and m be the number of words in the current remote communication process.

[0026] In one embodiment, the step of filtering each word segment by the TF-IDF value includes:

[0027] Each word segment that did not appear in the current remote communication process but appeared in the historical remote communication process is recorded as a historical word segment. The product of the TF-IDF value of each historical word segment and the preset attenuation coefficient is used as the relative importance of each historical word segment.

[0028] The TF-IDF values ​​of all words in the current remote communication process are sorted in descending order of relative importance with all historical words, and a preset number of words at the top of the sort are selected as the filtered words.

[0029] In one embodiment, the step of matching and recommending information to push out idle messages includes:

[0030] Calculate the similarity between the filtered word segments and all word segments of each idle message in the idle message push library, and push the idle message with the highest similarity.

[0031] This application has at least the following beneficial effects:

[0032] This application performs text recognition and word segmentation on information during each remote communication process; it can accurately extract key information from the communication content, laying the foundation for subsequent sentiment analysis and feature extraction. Based on the sentiment tendency of each word in each occurrence, the word polarity of each word in each occurrence is obtained; by analyzing the sentiment tendency of each word, the sentiment polarity of each word can be quantified; then, combined with the word polarity of all occurrences of each word in each remote communication process, the information tendency of each word in each remote communication process is determined; the determination of the information tendency helps to understand the overall interest change trend of the communication content and improves the accuracy of key information extraction during remote communication; the difference in the information tendency of each word in each remote communication process relative to the first remote communication is analyzed to determine the word tendency change rate of each word in each remote communication process; the word tendency change rate dynamically reflects the diachronic evolution of word sentiment expression by comparing the difference in information tendency between the current and the first communication; by using the importance in time, the word tendency change rate is weighted to obtain the word temporal interference rate of each word in each remote communication process; the word temporal interference rate strengthens the timeliness influence of recent communication data and weakens historical noise interference, making trend analysis more effective. This approach better aligns with real-world scenario changes. It analyzes the temporal interference rate trends of each word segment during all previous remote communication processes, determining the trend change of each word segment during the current remote communication. This trend change identifies the potential evolution direction of user communication interests, providing a basis for dynamically adjusting recommendation strategies. Based on the distribution of information tendency of each word segment during the current remote communication and combined with the trend change, the weight coefficient of each word segment during the current remote communication is obtained. The TF value in the TF-IDF algorithm is corrected using these weight coefficients, and the TF-IDF value of each word segment during the current remote communication is obtained using the corrected TF-IDF algorithm. By integrating historical tendency distribution and real-time trend, the TF value simultaneously reflects the global importance and local sentiment mutation characteristics of words, improving the distinguishability of keywords and solving the defect of traditional TF-IDF algorithms that rely solely on word frequency while ignoring semantic tendency. Each word segment is filtered using the TF-IDF value and matched with recommendation information to push hang-up messages, emphasizing the importance of the user's current communication content for the hang-up message push, reducing interference from invalid information, and improving the accuracy and applicability of the hang-up message push. Attached Figure Description

[0033] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 A block diagram of an adaptive push system for hang-up messages based on remote communication provided in this application;

[0035] Figure 2 Block diagram of the communication feature extraction module;

[0036] Figure 3 This is a block diagram of the AFK message push module. Detailed Implementation

[0037] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a remote communication-based adaptive push system for hang-up messages proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0039] The following description, in conjunction with the accompanying drawings, details a specific solution for an adaptive push system for hang-up messages based on remote communication provided in this application.

[0040] One embodiment of this application provides an adaptive push system for idle messages based on remote communication. Specifically, it provides the following adaptive push system for idle messages based on remote communication. Please refer to [link to relevant documentation]. Figure 1 The system includes: a communication data processing module 101, a communication feature extraction module 102, and a hang-up message push module 103.

[0041] The communication data processing module 101 is used to perform text recognition and word segmentation on the information in each remote communication process; and to obtain the word polarity of each word based on the sentiment tendency of each word in each occurrence.

[0042] This embodiment obtains the text data of each remote communication process by utilizing the background database of the remote communication system, and uses the jieba word segmentation tool to segment the text of each remote communication process to obtain the word segments of each remote communication process.

[0043] It should be noted that if the user uses voice signals during remote communication, the voice signals are converted into text through voice recognition. The use of the background database of the remote communication to obtain the text data of each remote communication process and the voice recognition are both existing well-known technologies, and will not be described in detail in this embodiment.

[0044] For each word segment in each remote communication process, the sentiment tendency of each word segment is analyzed to obtain the word polarity of each word segment. Specifically, this embodiment uses a hybrid model based on Transformer encoder and Long Short-Term Memory (LSTM) network to classify each word segment according to sentiment. The word polarity of words belonging to positive words is set to 1, the word polarity of words belonging to negative words is set to -1, and the word polarity of words that do not belong to either positive or negative words is set to 0. The hybrid model based on Transformer encoder and Long Short-Term Memory (LSTM) network is a well-known existing technology, and the specific process is not described in detail. Implementers can choose other existing feasible sentiment classification models, and this embodiment does not impose any restrictions.

[0045] The communication feature extraction module 102 (1) is used to combine the word polarity of each word in each remote communication process to determine the information tendency of each word in each remote communication process; analyze the difference of the information tendency of each word in each remote communication process relative to the first remote communication, and determine the word tendency change rate of each word in each remote communication process.

[0046] It should be understood that during remote communication, the more interested a user is in the information, the more frequently the information is repeated, and the more likely the word segmentation of the text information is to be positive after sentiment analysis. Therefore, when recommending hang-up messages to users, priority should be given to information that has been determined to be positive and has a high repetition frequency through sentiment analysis. This not only accurately meets user needs but also improves the user experience. Therefore, this embodiment calculates the information tendency of each word segmentation during each remote communication process based on the word polarity of all occurrences of each word segmentation. The specific calculation method is as follows:

[0047] In the formula, Hq i Cs represents the information tendency of the i-th word segment during each remote communication process. i Yu represents the number of times the i-th word appears during each remote communication process. i,j Let represent the polarity of the word that appears j times in the i-th word segment during each remote communication process.

[0048] In the analysis of information from remote user communications, high-frequency word segmentation is more effective in reflecting user interests; the higher the frequency of word segmentation, the greater the information attention it receives. This means that the information tendency calculated through these high-frequency word segments has higher reference value. When analyzing user behavior for information tendency, the positive tendency of high-frequency word segmentation can effectively improve the accuracy of message recommendations.

[0049] This embodiment obtains each word segment in each remote communication process of the user, calculates the information tendency of each word segment in each remote communication process, and obtains each word segment in the current remote communication process and all previous remote communication processes. Taking the i-th word segment as an example, if the i-th word segment does not appear in a remote communication process, the information tendency of the i-th word segment in that remote communication process is set to 0. Therefore, the information tendency of the i-th word segment in all remote communication processes can be obtained. The characteristics of the change of the information tendency of the i-th word segment over time are analyzed. If the information tendency of the word segment continuously increases and the rate of increase is faster, it indicates that the user is more interested in the matching information of the word segment.

[0050] Based on the above analysis, this embodiment calculates the word segmentation tendency change rate of each word segmentation during each remote communication process. The specific calculation method is as follows:

[0051] In the formula, Zt i,k Po represents the rate of change in segmentation tendency of the i-th segment during the k-th remote communication process. i,k Po represents the information tendency during the k-th remote communication process of the i-th word segment. i,1 τ represents the information tendency during the first remote communication process of the i-th word segment. τ is a preset value greater than 0 to avoid the denominator being 0. In this embodiment, τ = 0.01. Implementers can set it according to the actual situation. This embodiment does not impose any restrictions here.

[0052] It should be understood that during remote communication, the rate of change in information preference can effectively reflect the changing trends of user interests. When the information preference of a segmented word increases continuously over time, and the faster the rate of increase, the higher the user's interest in the associated messages. This means that message recommendations based on these high-increment-rate segmented words have greater reference value. When conducting interest analysis on user communication data, high-increment-rate changes in preference can effectively improve the accuracy of recommendations, thereby better meeting users' personalized needs for personalized messages.

[0053] (2) By using the importance in time, the word segmentation tendency change rate is weighted to obtain the word segmentation temporal interference rate in each remote communication process; the change trend of the word segmentation temporal interference rate in all previous remote communication processes is analyzed to determine the trend change amount of each word in the current remote communication process.

[0054] However, user preferences and interests can change dynamically over time. Therefore, information closer to the current remote communication process is more important for assessing user interests than information from historical remote communications. Time sensitivity is particularly important in precise recommendations because user interests can change significantly over time. For example, short-term user behavior may reflect immediate interests more closely, while historical data may contain outdated preferences. Therefore, recommendation information needs to rely more heavily on recent data to capture trends in user interests.

[0055] Based on the above analysis, this embodiment calculates the segmentation temporal interference rate of each segmentation in each remote communication process. The specific calculation process is as follows:

[0056] In the formula, Zy i,k Zt represents the temporal interference rate of the i-th segment in the k-th remote communication process, where k is the sequence number of the remote communication process. i,k Let k be the rate of change of segmentation tendency of the i-th segmentation during the k-th remote communication process. n Let N be the sequence number of the nth remote communication process, and N be the number of all previous and current remote communication processes.

[0057] It should be noted that the sequence number of the remote communication process is the chronological order of the remote communication process. For example, the sequence number of the first remote communication process is 1, the sequence number of the second remote communication process is 2, the sequence number of the third remote communication process is 3, and so on.

[0058] It should be understood that this embodiment assigns a time weight to each word segmentation tendency change rate to emphasize information closer to the current moment, which can more accurately capture users' recent interests and improve the accuracy of offline message recommendations.

[0059] Furthermore, to measure the changing trend of the temporal interference rate of each word in all previous remote communication processes, this embodiment uses the least squares method to obtain the fitted straight line of the temporal interference rate of each word in all previous remote communication processes in chronological order, and combines it with the numerical distribution of the temporal interference rate of each word in all remote communication processes to calculate the trend change of each word in the current remote communication process. The specific calculation method is as follows:

[0060] In the formula, A i Zy represents the trend change of the i-th word segment during the current remote communication process. i,k Let a be the segmentation temporal interference rate of the i-th segment in the k-th remote communication process, and N be the number of all previous and current remote communication processes. iThe slope of the fitted line for the i-th word segment in the current remote communication process is given by , and sign() is the sign function, which takes the value of 1 when the value is positive, -1 when the value is negative, and 0 when the value is 0.

[0061] It should be understood that during remote communication, the temporal interference rate of each segmented word over time can effectively reflect the direction and intensity of changes in user interest. By calculating the trend change of segmented words, the dynamic changes in user interest can be captured more accurately. When the trend change of segmented words is positive, it indicates that the user's interest in the information is increasing; when the trend change of segmented words is negative, it indicates that the user's interest in the information is decreasing; and when the trend change of segmented words is zero, it indicates that the user's interest is relatively stable. The trend change of segmented words can provide more accurate dynamics of user interest for message recommendation, thereby optimizing the recommendation effect and better meeting user needs. The communication feature extraction module block diagram is as follows. Figure 2 As shown.

[0062] Hang-up message push module 103, (1) Based on the distribution of the information tendency of each word in the current remote communication process in the historical remote communication process, combined with the trend change, obtain the weight coefficient of each word in the current remote communication process; correct the TF value in the TF-IDF algorithm through the weight coefficient, and obtain the TF-IDF value of each word in the current remote communication process using the corrected TF-IDF algorithm.

[0063] To emphasize the importance of information during the user's current remote communication process and highlight the user's current needs, this embodiment focuses on analyzing the numerical distribution of information tendency of each word in the current and previous remote communication processes. Combining this trend change, the weight coefficient of each word in the current remote communication process is calculated. The specific calculation method is as follows:

[0064] In the formula, W i Po represents the weight coefficient of the i-th word segmentation in the current remote communication process. i,k Let A be the information tendency in the k-th remote communication process of the i-th word segmentation, and N be the number of all remote communication processes, including the current one and those preceding it. i sig() represents the trend change of the i-th word segment during the current remote communication process, and sig() is the sigmoid normalization function.

[0065] It should be understood that when the information tendency of a word segment is very large in all remote communication processes, it indicates that the user is more interested in the recommended information corresponding to that word segment. The larger the weight coefficient of the obtained word segment, the higher the importance of the corresponding word segment. In subsequent message pushes, more consideration should be given to the relevance to that word segment, so that the weight coefficient of the word segment during the user's remote communication process is calculated more accurately, and the recommended message to the user is more precise.

[0066] Therefore, this embodiment modifies the calculation of the TF value in the TF-IDF algorithm, specifically as follows:

[0067] In the formula, tf(t) is the TF value after the t-th word segmentation correction in the current remote communication process, and W t Let f(t) be the weight coefficient of the t-th word in the current remote communication process, f(t) be the frequency of the t-th word in the current remote communication process, and m be the number of words in the current remote communication process.

[0068] Then, combining the word frequency weights of the word segments in the current remote communication process, the TF-IDF value of each word segment in the current remote communication process is calculated using the TF-IDF algorithm, reflecting the importance of each word segment in the current remote communication process. The calculation of the TF-IDF algorithm is a well-known technique, and the specific process will not be elaborated upon.

[0069] (2) Filter each word by the TF-IDF value and match it with the recommendation information to push the hang-up message.

[0070] Each word segment that did not appear in the current remote communication process but appeared in the historical remote communication process is recorded as a historical word segment. The product of the TF-IDF value of each historical word segment and the preset attenuation coefficient is used as the relative importance of each historical word segment. The purpose is to reduce the importance of word segments in the historical remote communication process for the recommendation of hang-up messages, so that the hang-up message adaptive push system will give more consideration to the word segments in the user's current remote communication process.

[0071] It should be noted that the attenuation coefficient ranges from [0.7 to 0.95]. In this embodiment, the attenuation coefficient is 0.9. Implementers can set the value within the range according to the actual situation. This embodiment does not impose any restrictions on this value.

[0072] The TF-IDF values ​​of all words in the current remote communication process are sorted in descending order of relative importance with all historical words. At the same time, each hang-up message in the hang-up message push library is also segmented using the jieba word segmentation tool. All words of each hang-up message and each word after the descending order are converted into word vectors. The similarity between the word vectors of all words of each hang-up message and the word vectors of the first preset number of words after the descending order is calculated. Specifically, the number of words in the first preset number of words is equal to the number of words in each hang-up message. For example, if there is a hang-up message with 5 words, then when calculating the similarity with that hang-up message, the first 5 words after the descending order are selected for calculation.

[0073] It should be noted that the cosine similarity is used to calculate the similarity between word vectors in this embodiment. Implementers may choose other feasible similarity calculation methods, such as Pearson correlation coefficient.

[0074] Finally, the most similar idle message from the idle message push library is pushed to the user, completing adaptive push of idle messages and improving the accuracy of idle message push. The idle message push module block diagram is as follows: Figure 3 As shown.

[0075] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0076] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0077] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them; modifications to the technical solutions described in the foregoing embodiments, or equivalent substitutions of some of the technical features, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A remote communication-based adaptive push system for hang-up messages, characterized in that, The system includes: The communication data processing module is used to perform text recognition and word segmentation on the information during each remote communication process; and to obtain the word polarity of each word segmentation based on the sentiment tendency of each word segmentation. The communication feature extraction module is used to determine the information tendency of each word in each remote communication process by combining the word polarity of all occurrences of each word in each remote communication process; analyze the difference of the information tendency of each word in each remote communication process relative to the first remote communication, and determine the word tendency change rate of each word in each remote communication process; By weighting the word segmentation tendency change rate according to the importance in time, the word segmentation temporal interference rate of each word in each remote communication process is obtained; the change trend of the word segmentation temporal interference rate of each word in all previous remote communication processes is analyzed to determine the trend change amount of each word in the current remote communication process. The message push module for hanging up (or ignoring communication) obtains the weight coefficient of each word segment in the current remote communication process based on the distribution of information tendency in the historical remote communication process and the trend change amount. It then corrects the TF value in the TF-IDF algorithm using the weight coefficient and uses the corrected TF-IDF algorithm to obtain the TF-IDF value of each word segment in the current remote communication process. Each word segment is filtered using the TF-IDF value and matched with the recommendation information to push the AFK message; The determination of the word segmentation temporal interference rate includes: Calculate the sum of the sequence numbers of the current and all previous remote communication processes, calculate the ratio of the sequence number of each remote communication process to the sum, and the word segmentation temporal interference rate is the product of the word segmentation tendency change rate and the ratio; The determination of the weighting coefficients includes: The normalized result of the mean of the information tendency of each word in the current remote communication process in all previous remote communication processes is calculated, and the weight coefficient is the product of the normalized result and the normalized result of the trend change.

2. The adaptive push system for hang-up messages based on remote communication as described in claim 1, characterized in that, The process of obtaining the word polarity for each occurrence of each word segment includes: The semantic analysis algorithm is used to obtain the sentiment classification results of each word segmentation at each occurrence. The word polarity of the word segmentation belonging to positive words is set to 1, the word polarity of the word segmentation belonging to negative words is set to -1, and the word polarity of the word segmentation belonging to neither positive nor negative words is set to 0.

3. The adaptive push system for hang-up messages based on remote communication as described in claim 1, characterized in that, The determination of the information tendency includes: For each remote communication, the sum of word polarity for all occurrences of each word segment is calculated. The information tendency is positively correlated with the sum and the number of occurrences of each word segment.

4. The adaptive push system for hang-up messages based on remote communication as described in claim 1, characterized in that, The rate of change of word segmentation tendency is positively correlated with the difference between each word segmentation and negatively correlated with the amount of information tendency of each word segmentation during the first remote communication process.

5. The adaptive push system for hang-up messages based on remote communication as described in claim 1, characterized in that, The determination of the trend change includes: Obtain the slope of the fitted straight line of the temporal interference rate of any segmentation in the current and all previous remote communication processes; Calculate the sign result of the sum of the temporal interference rates of any segment in the current and all previous remote communication processes for any segment, and the trend change of any segment in the current remote communication process is the product of the slope and the sign result.

6. The adaptive push system for hang-up messages based on remote communication as described in claim 1, characterized in that, The method for correcting the TF value in the TF-IDF algorithm using the weighting coefficients is as follows: In the formula, This represents the TF value after correction for the t-th word segmentation during the current remote communication process. Let be the weight coefficient of the t-th word segment in the current remote communication process. Let be the frequency of the t-th word segmentation during the current remote communication process. This represents the number of words segmented during the current remote communication process.

7. The adaptive push system for hang-up messages based on remote communication as described in claim 1, characterized in that, The step of filtering each word segment by the TF-IDF value includes: Each word segment that did not appear in the current remote communication process but appeared in the historical remote communication process is recorded as a historical word segment. The product of the TF-IDF value of each historical word segment and the preset attenuation coefficient is used as the relative importance of each historical word segment. The TF-IDF values ​​of all words in the current remote communication process are sorted in descending order of relative importance with all historical words, and a preset number of words at the top of the sort are selected as the filtered words.

8. The adaptive push system for hang-up messages based on remote communication as described in claim 7, characterized in that, The above is combined with the recommended information to push out AFK messages, including: Calculate the similarity between the filtered word segments and all word segments of each idle message in the idle message push library, and push the idle message with the highest similarity.

Images