Technical Field

This disclosure relates to information and communication technology, and in particular, to a system and a method for message distribution.

Background

With the vigorous development of the Internet and mobile networks, electronic products (such as a mobile phone, a tablet computer, a desktop computer, a notebook, and a smart appliance) having a network access function have been widely used by the public. In addition, information is massive on networks and can be conveniently obtained. In order to attract people’s attention, some APPs or platforms always distribute messages to user terminals by pushing notifications.

Current push notifications are sent automatically or manually to bring back users to the platform and build/increase their affinity towards the platform. The current System is not personalized, and it is limited to the streamers that the user follows, the high-performing streamers, or offers/promotions.

However, notifications with inappropriate content at the wrong time do not attract people’s attention at all, sometimes it makes the user feel annoyed. Therefore, personalized notifications with appropriate content at an appropriate time is a very important issue.

Summary

An embodiment of sub _j ect application relates to a message distribution system, including a collecting unit and an evaluating unit. The collecting unit is configured to collect feedback on a message from a user terminal. Hie evaluating unit is connected to the collecting unit and is configured to evaluate a preference for the message of the user terminal according to the feedback. The feedback includes a response action feedback and/or an influence action feedback. The response action feedback includes an operation to the message by the user terminal. The influence action feedback includes a behavior followed by the operation. Another embodiment of subject application relates to a message distribution method, including collecting feedback on a message from a user terminal; and evaluating a preference for the message of the user terminal according to the feedback. The feedback includes a response action feedback and an influence action feedback. The response action feedback includes an operation to the message by the user terminal. The influence action feedback includes a behavior followed by the operation.

The present disclosure increases the discoverability of the content that users are looking for, and further improves the satisfaction of the message to the user and optimizes the user experience.

Brief description of the drawings

FIG. 1 is a schematic configuration of a communication system 1 according to some embodiments of the subject application;

FIG. 2 is a schematic block diagram of the message distribution system 100 according to some embodiments of the subject application;

FIG. 3 is a schematic configuration of a communication system G according to some embodiments of the subject application;

FIG. 4 is a schematic block diagram of the message distribution system 100’ according to some embodiments of tire subject application.

FIG. 5 is a schematic block diagram of the preference evaluating element 124 according to some embodiments of the subject application.

Detailed Description

FIG. 1 is a schematic configuration of a communication system 1 according to some embodiments of the subject application. The communication system 1 includes a user terminal 10 and a server 20. The user terminals 10 and the server 20 are connected via a network 90, which is the internet, for example. Server 20 includes a message distribution system 100. The message distribution system 100 is configured to send a message M to the user terminal 10 and receive feedback F on the message M from the user terminal 10. in some embodiments, message M may be a text message, mail, chat bubble, push message, push notification, and the like. In some embodiments, the contents of the message M may be text, picture, sound, audio, video, live streaming, podcasting, sales, and the like, in some embodiments, the user terminal 10 may be a device a user uses. The device may be electronic products. The user may be referred to as a viewer, streamer, podcaster, audience, listener or the like. The communication system 1 may include a plurality of user terminals 10, and the user terminal 10 are shown in FIG. 1 and FIG. 2 for simplification.

FIG. 2 is a schematic block diagram of the message distribution system 100 according to some embodiments of the subject application. The message distribution system 100 includes a collecting unit 110, an evaluating unit 120, and a pushing unit 130. The collecting unit 110 is configured to collect a feedback F on a message M from a user terminal 10. The evaluating unit 120 is connected to the collecting unit 110 and configured to evaluate a preference P for the message M of the user terminal 10 according to the feedback F. The pushing unit 130 is connected to the evaluating unit 120 and configured to push the message M to the user terminal 10.

As shown in FIG. 2, the pushing unit 130 pushes a message M to the user terminal 10 and the collecting unit 110 collects the feedback F on the message M from the user terminal 10. The collecting unit 1 10 transmits the feedback F on the message M to the evaluating unit 120 and the evaluating unit 120 evaluates a preference P for the message M of the user terminal 10 according to the feedback F. Then, the pushing unit 130 receives the preference P from the evaluating unit 120. If the user terminal 10 has the preference P for the message M, the pushing unit 130 pushes the message M to the user terminal 10. On the other hand, If the user terminal 10 has no preference P for the message M, the pushing unit 130 stops pushing the message M to the user terminal 10.

The feedback F may refer to an action the user terminal 10 takes against the message M. in some embodiments, the feedback F may include a response action feedback Fr and an influence action feedback Fi. The response action feedback Fr may refer to an operation to the message M operated by the user terminal 10. Tire influence action feedback Fi may refer to a behavior followed by the operation. Once the user terminal 10 took actions against the message M, the user terminal 10 transmits data to the message distribution system 100. Different data corresponds to different types of action. The evaluating unit 120 evaluates the preference P for the message M according to different types of data. in some embodiments, the operation may include clicking, swiping, deleting, ignoring, silencing, turning off notifications, expanding, minimizing, or other operations the user terminal 10 may operate to the message M, and the combination of above. For example, the operation may include clicking the message M or swiping to delete the message M. In some embodiments, the operation may also include ignoring the message M, turning off the pushing notification function of the device, or unsubscribing the push notification of the message M or the like. In some embodiments, the message M may be expanded from a simplified message M to a full message M. The operation may further include expanding and clicking the message M, expanding and swiping to delete the message M and so on.

In some embodiments, the evaluating unit 120 evaluates the preference P of the user terminal 10 according to the operation. For example, if the user terminal 10 operates the message M and opens the APP, it shows that the user terminal 10 has the preference P for the message M. On the other hand, if the user terminal 10 operates the message but does not open the APP, it shows that the user terminal 10 has no preference P for the message M. For example, the evaluating unit 120 evaluates that the user terminal 10 has a preference P for the message M if the user terminal 10 clicks the message M, and evaluates that the user terminal 10 has no preference P for the message M if the user terminal 10 swipes to delete the message M. Once the user terminal 10 operates the message M with an operation, the user terminal 10 transmits data to the message distribution system 100. Different data corresponds to different types of operations. Hie evaluating unit 120 evaluates the preference P for the message M according to different types of data.

In some embodiments, the preference P may include the preference P for the simplified message M and the preference P for the full message M. For example, if the user terminal 10 expands the simplified message M to a full message M and clicks the message M, it shows that the user terminal 10 has the preference P for both the simplified and full message M. On the other hand, if the user terminal 10 expands the simplified message M to a full message M but does not click the message M, it show's that the user terminal 10 has the preference P for the simplified message M but has no preference tor the full message M. If the user terminal 10 neither expands nor clicks the message M, it shows that the user terminal 10 has no preference P for the simplified message M and tire full message M at all.

In some embodiments, the behavior may include being idle, searching, visiting, following, purchasing, sending, watching, commenting, supporting, marking, collecting, claiming, or tire action the user may perform in the APP, and the combination of above. In other words, the behavior may be an action the user terminal 10 takes or a state the user terminal 10 is in after operating the message M. For example, the user terminal 10 may be idle at the front page for a moment and close the app. In some embodiments, the user terminal 10 may search for streamers or events, or visiting the streamer’s profile or explore sections, in some embodiments, the user terminal 10 may enter a live room and chat with the streamer, or send gifts to the streamer. In some embodiments, the user terminal 10 may comment on articles or events. In some embodiments, the user terminal 10 may purchase gifts, points, or membership in the online store. In some embodiments, the user terminal 10 may follow specific streamers, support a streamer, or marking the user as loyal fans or favorite streamers, and so on.

In some embodiments, the evaluating unit 120 evaluates the preference P of the user terminal 10 according to the behavior, if the user terminal 10 spends time and takes some actions in the app, it shows that the user terminal 10 has a preference P for the message M. On the other hand, if the user terminal 10 is being idle without doing anything and closes the app, it show's that the user terminal 10 has no preference P for the message M. For example, if the user terminal 10 opens the APP and searches for a streamer, the evaluating unit 120 evaluates that the user terminal 10 has the preference P for the message M. On the other hand, although the user terminal 10 opens the APP but exits the APP soon without watching any stream, the evaluating unit 120 evaluates that the user terminal 10 has no preference P for the message M. Once the user terminal 10 performs a behavior after operating message M, the user terminal 10 transmits data to the message distribution system 100. Different data corresponds to different types of behavior. The evaluating unit 120 evaluates the preference P for the message M according to different types of data

In some embodiments, the collecting unit 110 collects feedbacks F on messages M from the user terminal 10 dynamically. More specifically, the collecting unit 110 does not only collect the real-time feedback F, but the historical feedback F. in other words, the feedback F includes real-time feedback F and historical feedback F, More specifically, the response action feedback Fr further includes real-time response action feedback Fr and historical response action feedback Fr. Similarly, the influence action feedback Fi further includes real-time influence action feedback Fi and historical influence action feedback Fi.

As shown in FIG. 2, the evaluating unit 120 further includes a message analysis element 121, a frequency analysis element 122 and a relevance analysis element 123. The message analysis element 121 is configured to analyze, filter and/or screen a user message Mu the user terminal 10 left in the APP. The frequency analysis element 122 is configured to identify a frequency Freq of the user terminal 10’s feedback F on the message M. The relevance analysis element 123 is configured to analyze a relevance R between the message M and the feedback F.

In some embodiments, the message analysis element 121 is configured to identify the contents of the user message Mu left by the user terminal 10. The user message Mu may be articles, comments, chat bubbles, emoji and so on, and contents of the user message Mu may be text, picture, sound, audio, video message and so on. The message analysis element 121 may be configured to determine if the user message Mu (such as a piece of chat content) is positive, negative or neutral, in some embodiments, the message analysis element 121 has a keyword identification part, to identify a keyword in the user message Mu, and to evaluate the preference P based on the keyword. For example, the message analysis element 121 may identify positive contents such as "do", "good", "like", "love", "interesting", smile emoji and the like in a user message Mu, and evaluate that the user message Mu is positive. On the other hand, the message analysis element 121 may identify negative contents such as "not", "bad", "dislike", "boring", sad emoji and the like in a user message Mu, and evaluates that the user message Mu is negative. In some embodiments, if no positive or negative contents are identified, the message analysis element 121 may evaluate that the user message Mu is neutral. in some embodiments, the evaluating unit 120 evaluates the preference P of the user terminal 10 according to the user message Mu. For example, the evaluating unit 120 evaluates that the user terminal 10 has a preference P for the message M if the user terminal 10 comments on a streamer in a streaming room and the comments are positive, which means that the user terminal 10 likes the streamer. On the other hand, if the user terminal 10 comments on the streamer and the comments are negative, it shows that the user terminal 10 has no preference P for the message M. In some embodiments, if the user message Mu is neutral, the user message Mu does not contribute to the preference P. in some embodiments, the frequency analysis element 122 may identify the frequency Freq of the feedback F. More specifically, the frequency analysis element 122 may identify the frequency Freq of the response action feedback Fr and/or the influence action feedback Fi. The frequency analysis element 122 may he configured to calculate the frequency Freq of the user terminal 10’s operation and/or behavior, and determine if the frequency Freq is positive, negative or neutral . in some embodiments, the evaluating unit 120 evaluates the preference P of the user terminal 10 according to the frequency Freq from the frequency analysis element 122. If the user terminal 10 has a specific operation and/or behavior frequently, it shows that the user has the preference P for the message M. For example, the user terminal 10 watches a streamer every time after clicking the message M, so the frequency analysis element 122 analyzes that the frequency Freq is positive and the evaluating unit 120 evaluates that the user terminal 10 has a preference P for the message M. in some embodiments, the frequency Freq may be referred to as the number of times of a specific operation and/or behavior, if the number of times is higher, between or lower than a specific range, it shows that the frequency Freq is high, intermediate or low'. Once the evaluating unit 120 receives the data of the user terminal 10, the evaluating unit 120 evaluates the similarity of the types of data with respect to the message M. if there is similar or the same types of data with respect to the message M, the frequency analysis element 122 begins to calculate the frequency Freq. With different frequency Freq, it may contribute a different increment or decrement to the preference P.

In some embodiments, the frequency analysis element 122 may further detect the continuity of the operation and/or behavior. For example, if the user terminal 10 watched a streamer frequently such as every day, but the user terminal 10 did not watch the streamer in one day, the frequency analysis element 122 detects that the frequency Freq is not continuous, in this situation, the user terminal 10 may he busy and forget to watch the streamer, the pushing unit 130 may push the message M or a reminder message to the user terminal 10. However, if the user terminal 10 did not dick the message M and watch the streamer anymore, the frequency analysis element 122 analyzes that the frequency Freq is negative. On the other hand, if the user terminal 10 starts to dick the message M and watches the streamer, the frequency analysis element 122 may analyze that the frequency Freq is positive.

In some embodiments, the frequency analysis element 122 further detects the consistency of the operation and/or behavior. For example, if the user terminal 10 watches a streamer every' time when clicking the message M, but the user terminal 10 does not follow the streamer, the frequency analysis element 122 may detect the inconsistency between watching a streamer and following a streamer. In this situation, the pushing unit 130 may push a message M ⁵ related to the message M such as recommending to follow the streamer, giving free gifts for following the streamer, or the like. In some embodiments, the collecting unit 110 may further collect a feedback F’ on the message M’ and the evaluating unit 120 may further evaluate the preference P’ for the message M’ of the user terminal 10.

In some embodiments, the relevance analysis element 123 may he configured to identify the relevance R between the message M and the feedback F. More specifically, the relevance analysis element 123 is configured to identify the contents of the message M and the operation and/or the behavior of the user terminal 10 and determine the relevance R between the message M and the feedback F. The relevance analysis element 123 may be configured to determine whether the rele vance R is relevant or irrelevant.

In some embodiments, the evaluating unit 120 evaluates the preference P of the user terminal 10 according to the relevance R from the relevance analysis element 123. Sometimes, the user terminal 10’s behavior has nothing to do with the contents of the message M. For example, the message M may recommend the user terminal 10 to watch a streamer’s live streaming. Even though the user terminal 10 clicks the message M, the user terminal 10 watches another streamer instead of watching the recommended streamer, in this situation, the relevance R of the message M and the behavior is irrelevant and the feedback F may not contribute to the preference P. On the other hand, if the relevance R is relevant, the feedback F may contribute to the preference P. By means of that, the accuracy of the preference P can be improved. in some embodiments, if the relevance R is relevant, it may further be classified into positively relevant or negatively relevant. For example, the message M may recommend the user terminal 10 to watch a streamer’s live streaming. After clicking and watching the streamer, the user terminal 10 may subscribe or unsubscribe to the streamer, the relevance analysis element 123 may determine the relevance R as positively relevant or negatively relevant and then contribute an increment or decrement respectively to the preference P. in some embodiments, the feedback F may further include time information. The evaluating unit 120 further evaluates the preference P according to the time information. More specifically, the evaluating unit 120 further evaluates the preference P according to time information of the response action feedback Fr and an influence action feedback Fi. The time information may refer to a time period such as in the evening, in the morning, at noon, at night, on the weekday, on the weekend or on vacation. The time information may also refer to a time span of getting involved in the APP such as 1 minute, 10 minutes, 1 hour, or more. The time information may also refer to the time zone of the user terminal 10 such as the Japan time zone or the U.S. time zone. For example, the user terminal 10 tends to click the message M and watch live streaming at night, so the evaluating unit 120 evaluates that the user terminal 10 has the preference P for operating the message M and performing behavior at night, the user terminal 10 always swipes to delete the message M in the morning, so it shows that the user terminal 10 has no preference P tor the message M in the morning.

In some embodiments, the feedback F may further include location information. The evaluating unit 120 further evaluates the preference P according to the location information. More specifically, the evaluating unit 120 further evaluates the preference P according to location information of the response action feedback Fr and an influence action feedback Fi. Hie location information may refer to country, city, town, home, workplace, restaurant, or the like. The location information may also refer to transportation means such as automobile, bus, rail, ship, airplane, or the like. For example, the user terminal 10 clicks the message M and watches streaming home, so the evaluating unit 120 evaluates that the user terminal 10 has the preference P for operating the message M at home, the user terminal 10 always swipes to delete the message M in the workplace, so it shows that tire user terminal 10 has no preference P for the message M in the workplace. In some embodiments, the pushing unit 130 may further push the message M to other users with similar location information. For example, A large number of users in Taipei city has visited Arthur's profile, the pushing unit 130 may push the message M to other users in Taipei city to recommend them to watch the streamer.

In some embodiments, the feedback F may further include device information. The evaluating unit 120 further evaluates the preference P according to the device information. More specifically, the evaluating unit 12.0 further evaluates the preference P according to a device information of the response action feedback Fr and an influence action feedback Fi. The user terminal 10 may generate the APP by a device such as mobile phone, table, smart appliance, and the like. The device information may refer to a current mode of the device, such as in a turnoff mode, standby mode, idle mode, busy mode or the like. The device information may also refer to a mode the user terminal 10 sets such as do not disturb mode, quiet mode, silent mode or the like. For example, the user tends to swipe to delete the message while being in busy mode, so it shows that the user terminal 10 has no preference P for the message M while being in busy mode.

In some embodiments, the device information may also include an operation time from the message being read to the message being operated. The evaluating unit 120 may evaluate the preference P according to the operation time. In some embodiments, if the message M is operated right after the message M is read, it may show that the riser terminal 10 tends not to check the contents of the message M or even does not like any kind of message. For example, if the user terminal 10 clicks the message M in less than 5 second, it may show that the user terminal 10 tends not to check the message M and always clicks any kind of messages M, so the feedback F may not contribute to the preference, if the user terminal 10 swipes to delete the message M in less than 1 second without checking the contents of the message M, it may show that the user terminal 10 does not like to receive any kind of messages.

In some embodiments, the evaluating unit 120 may further evaluate the preference P according to the combination of the above information. For example, the user terminal 10 in New York city frequently clicks the message M and watches live streaming in the evening, the evaluating unit 120 evaluates that the user terminal 10 with time information of “in the evening” and with location information of “New' York City” has a preference P for the message M. In some embodiments, the combination of the above information may be decided according to practical need.

In some embodiments, the preference P may be evaluated according to the retention rate of the user terminal. The retention rate may refer to the rate the user terminal 10 comes back and opens the APP after receiving the message M. In some embodiments, the retention rate may be the number of times of opening the APP or the length of engaging in the APP during a period of time, in some embodiments, the period of time may be 3 days, 7 days, 14 days or the like. For example. If the number of times of opening the APP or length of engaging in the APP in the following 7 days is increased after receiving the message M, it shows that the user terminal 10 has the preference P for the message M.

In some embodiments, the preference P may also be evaluated according to other criteria such as push notification dick rate, push notification unsubscribe rate, session length on opening APP and getting involved in the APP through push notification, frequency of opening APP or the like. If the push notification click rate is increased, it shows that the riser terminal 10 has the preference P for the message M. If the push notification unsubscribe rate is increased, it shows that the user has no preference P for the message M. If the number of times of opening the APP by clicking the message M is increased, it shows that user terminal 10 has the preference P for the message M. If the time length of engaging in the APP is increased, it shows that the user terminal 10 has the preference P for the message M. If the frequency of opening APP is increased, it shows that the user terminal 10 has the preference P for the message M.

In some embodiments, the evaluating unit 120 may further include a preference evaluating element 124. The preference evaluating element 124 is configured to evaluate the preference P according to the feedback F by means of the criteria above. In some embodiments, the preference evaluating element 124 may evaluate the preference P by means of the criteria respectively or a combination of the criteria above. FIG. 5 is a schematic block diagram of the preference evaluating element 124 according to some embodiments of subject application. The preference evaluating element 124 includes a criteria calculator

1241, a criteria evaluator 1242, a criteria module 1243, and a weighting module 1244. The criteria calculator 1241 is configured to calculate the preference P according to the criteria or the combination of the criteria. The criteria evaluator

1242. is configured to evaluate the criteria and the corresponding weighting for each criterion in order to calculate the preference P. The criteria module 1243 is configured to store the criteria for calculating the preference P. The weighting module 1244 is configured to store the weightings for each criterion.

As shown in FIG. 5, the criteria evaluator 1242 retrieves the criteria and the corresponding weightings from the criteria module 1243 and the weighting module 1244. The criteria evaluator 1242 may evaluate the corresponding weighting for each criterion to determine suitable weighting for each criterion. After the weighting for each criterion is determined, the criteria evaluator 1242 may update the weighting module 1244. Then The criteria calculator 1241 may retrieve the criteria and the corresponding weightings to calculate the preference P. For example, the criteria evaluator 1242 may retrieve some criteria such as criterion Cl, criterion C2, and criterion C3 from the criteria module 1243. In some embodiments, the criterion Cl may be a push notification click rate, the criterion C2 may be session length on geting involved in the APP through push notification, and criterion C3 may be frequency of opening APP.

Once the criteria are determined, the criteria evaluator 1242 may evaluate the weighting for each criterion. For example, the criteria evaluator 1242 may evaluate the weighting Wl, weighting W2 and weighting W3 for criterion Cl, criterion C2 and criterion C3 respectively. In some embodiments, the total of the weighting Wl, W2 and W3 may be 1 and the weighting Wl, W2 and W3 may ^¬ be 0.3, 0.3 and 0.4 by default, for example, in some embodiments, the default values of weightings may be determined based on the practical need. In some embodiments, the weightings may vary' dynamically accordingly.

In some embodiments, the criteria evaluator 1242 may evaluate the weighting according to real-environment data from a real -environment database. As shown in FIG. 5, the criteria evaluator 1242 may retrieve the real- environment data from the real -environment database. In some embodiments, the real -environment data may include the real-environment data of the APP For example, the real-environment data may be the time length the user terminal 10 gets involved in the APP, or the total profit the user 10 contributes to the APP. More specifically, the profit the user 10 contributes may be the gift or reward the user terminal 10 bought, donated or received, or the business value of the APP. The criteria evaluator 1242 may adjust the weightings for each criterion if the real-environment data is decreased or is not increased as expected. The criteria evaluator 1242 may further trigger the criteria calculator 1241 to generate a fonnuia for calculating the preference P. The formula may be like the following:

Preference P W i * Cl + W2* C2 + W3*C3. (1)

In some embodiments, the preference evaluating element 124 may further include a training pairing unit 1245 and a model generator 12.46, The criteria evaluator 1242 may evaluate the criteria C1-C3, adjust the weighting W1-W3 and trigger the criteria calculator 1241 to update the training pairing unit 1245. The training pairing unit 1245 is configured to train and pair the preference P updated by the criteria calculator 1241 and the feedback F. The model generator 1246 is configured to generate a model for the preference P with respect to the feedback F. In some embodiments, the training pairing unit 1245 ingests the feedback F, trains the preference P updated by the criteria calculator 1241, and pairs the preference P for the feedback F. The model generator 1246 retrieves the preference P and the feedback F and generates a model for the preference P and the feedback F. In some embodiments, the weightings vary dynamically with respect to the real -environment data, for example. More specifically, the criteria calculator 1241, the criteria evaluator 1242, the training pairing unit 1245 and the model generator 1246 works dynamically to generate a suitable model for the preference P with respect to the feedback F. Eventually, the well-trained models came out from the model generator 1246. According to the present invention, a more accurate preference P for the feedback F may be evaluated and the user satisfaction for the message M such as push notification may be improved. Furthermore, the click rale of the message, the click rate, involving time, and the business value of the APP may be improved accordingly. in some embodiments, the evaluating unit 120 may further be configured to identify if there is a change in a user’s operation and/or behavior resulting in increased or decreased retention rate. In this situation, the pushing unit 130 may orient the user terminal 10 toward certain feedback by pushing a corresponding message in order to increase the retention rate or other criteria. in some embodiments, the message distribution system 100 may be behavior-oriented. More specifically, the message distribution system 100 may orient the user terminal 10 toward a specific behavior by pushing an orientation message Mo with positive preference Po. For example, the message distribution system 100 may orient the riser terminal 10 toward watching a streamer’s live streaming, in order to orient the user terminal 10 toward watching the streamer, the pushing unit 130 pushes an orientation message Mo to the user terminal 10. The orientation message Mo has a positive preference Po from most of the users, or from the same group as the riser terminal 10. The pushing unit 130 pushes the orientation message Mo to the user terminal 10 to orient the user terminal 10 toward a similar operation and/or behavior. In some embodiments, the collecting unit 110 may further collect a feedback Fo on the orientation message Mo and the evaluating unit 120 may further evaluate the preference Po for the orientation message Mo of the user terminal 10. in some embodiments, the evaluating unit 120 evaluates the preference P for the message M and determines whether the pushing unit 130 pushes the message M to the user terminal 10 or not. In some embodiments, the pushing unit 130 may further push a message M’ related to the message M based on the preference P. For example, the message M recommends John watching Paul’s streaming, John clicks the streaming and gifts Paul during the streaming. Next time when Paul is streaming again then the message M may he pushed to John. Furthermore, A message M ^' about free gifts or discount information may be pushed to John in the future. In some embodiments, the pushing unit 130 may further recommend a recommendation message Mr to the user terminal 10 according to an operation and/or behavior in the APP. For example, the user terminal 10 opens the APP and scrolls down the feed but exits without watching any stream, the pushing unit 130 may push a recommendation message Mr to recommend the user terminal 10 another stream or discount information. in some embodiments, the preference P may be expressed as a percentage and a default of the preference P may be 50%. The feedback F may be classified as positive, negative, or neutral feedback F. More specifically, the operation may be classified as a positive, negative and neutral operation, the behavior may be classified as positive, negative and neutral behavior. Once a positive, negative or neutral operation is operated, the preference P may be increased by an increment, decreased by a decrement or constant. Similarly, once a positive, negative or neutral behavior is performed, the preference P may be increased by an increment, decreased by a decrement or constant.

If the preference P is higher than a threshold Tli, the evaluating unit 120 evaluates the user terminal 10 has a preference P, and the pushing unit 130 may further push the message M to the user terminal 10 in the future. On the other hand, If the preference P is lower than a threshold Ti, the e valuating unit 120 evaluates the user terminal 10 has no preference P, and the pushing unit 130 may stop pushing the message M to the user terminal 10. Moreover, If the preference P is within the threshold Th and the threshold Ti, the pushing unit 130 and the collecting unit 110 may keep pushing the message M and collecting the feedback F to the evaluating unit 120 may keep evaluating the preference P tor the message M of the user terminal 10 to determine if the user terminal 10 has a preference P for the message M. in some embodiments, the increment and the decrement may be 3%, 5%, 10% or the like. The threshold TI may be 20%, 30%, 40% or the like. The threshold Th may be 60%, 70%, 80% or the like. In some embodiments, the default of the preference P, the increment, decrement, the threshold TI or the threshold Th may be determined depending on practical needs.

FIG. 3 is a schematic configuration of a communication system G according to some embodiments of the subject application. FIG. 4 is a schematic block diagram of the message distribution system 100’ according to some embodiments of the subject application. To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures. As shown in FIG. 3, The communication system G includes user terminal 10.4, user terminal 10B, user terminal IOC, and a server 20. The user terminals 10A-10C and the server 20 are connected via a network 90, which is the Internet, for example. Server 20 includes a message distribution system 1002 The communication system G may include a plurality of users, and the user terminal 10A, user terminal 10B and user terminal IOC are shown in FIG. 3 for simplification.

The message distribution system 100’ includes a collecting unit 110, an evaluating unit 120, a pushing unit 130 and a comparing unit 140. The collecting unit 110 is configured to collect feedback FI, F2 and F3 on a message M from a user terminal 10A, J OB and IOC respectively. The evaluating unit 120 is connected to the collecting unit 110 and configured to evaluate a preference Pi, P2 and P3 for the message M of the user terminal 10A, 10B and IOC according to the feedback FI, F2 and F3.

As shown in FIG. 4, the comparing unit 140 is connected to the collecting unit 110 and the evaluating unit 120 respectively, and is configured to compare the feedback FI, F2 and/or F3 from the collecting unit 110 and the preference PI, P2 and/or P3 to the feedback FL F2 and/or F3 from the evaluating unit 120. More specifically, the comparing unit 140 compares the feedback F collected from a plurality of users and the preference P to the feedback F evaluated by the evaluating unit 120. The pushing unit 130 is connected to the comparing unit 140 and configured to push the message M to the user terminal 10. in some embodiments, the comparing unit 140 compares the feedback FI and the preference Pi from one specific user such as the user terminal 10A. More specifically, the comparing unit 140 compares whether the preference Pi is consistent with the feedback FI. For example, if the user terminal 10A clicks the message M and purchases a premium membership and the evaluating unit 120 evaluates that the user terminal 10A has a preference P for the message M. Clicking message M may be referred to as a positive operation and Purchasing a premium membership may be referred to as a positive behavior. Therefore, the comparing unit 140 compares the feedback F and the preference P and confirms that the preference P is consistent with the feedback F. in some embodiments, the comparing unit 140 may be configured to detect whether the preference PI for the message M of the user terminal 10A has changed. The comparing unit 140 may compare the real-time feedback FI with historical feedback FI from the user terminal 10A, If the real-time feedback FI is consistent with the historical feedback FI, it shows that the preference PI does not change. On the other hand, if the real-time feedback FI is not consistent with the historical feedback FI, it may be predicted that the preference PI tor the message may be changed. For example, the user terminal 10A used to click the message M in the past, so it shows that the user terminal iOA had preference P for the message M. However, the user terminal 10A is swiping to delete the message M in real-time, so it shows that the user terminal 10A has no preference Pi for the message M. Therefore, the comparing unit 140 compares the real-time feedback FI, which is swiping to delete the message M, with the historical feedback FI, which is clicking the message M, and detects that the preference PI for the message M has changed.

In some embodiments, the comparing unit 140 compares the feedback FI, F2 and F3 on the message M from a plurality of users such as the user terminal 10A, 10B and IOC. if the feedback FI is similar to the feedback F2, the user terminal 10A and the user terminal 10B may be classified into the same group. If the feedback FI is different from the feedback F3, the user terminal 10A and the user terminal IOC may be classified into different groups. For example, if the user terminal 10A clicks the message M and follows a streamer and the user terminal 10B also does so, it shows that the user terminal 10A and the user terminal I0B have similar preference for the message M. In this situation, the user terminal 10A and user terminal 10B may be classified into the same group. On the other hand, if the user terminal IOC swipes and deletes the message M, it shows that the user terminal 10A and the user terminal IOC have different preferences for the message M and the user terminal 10A and the user terminal IOC may be classified into different groups. Here, being similar may be referred to having the same operation and/or behavior, and being different may be referred to having different operations and/or behaviors.

As shown in FIG. 4, the pushing unit 130 may push the message MI, M2 and M3 to tire user terminal 10A, 10B and IOC respectively. The collecting unit110 may collect feedback FI, F2 and F3 on the message Ml, M2 and M3 from a user terminal 10A, 10B and 10C respectively. The evaluating unit 120 may evaluate a preference PI, P2 and P3 for the message Ml, M2 and M3 of the user terminal 10A, 10B and IOC according to the feedback FI, F2 and F3. If the user terminal 10A has a preference PI for the message Ml according to the feedback FI, the message Ml may further be recommended to the user in the same group such as the user terminal 10B. In some embodiments, the collecting unit 110 may further collect feedback on the message Ml to the user terminal 10B and the evaluating unit 120 may further evaluate the preference for the message Ml of the user terminal 10B. Furthermore, a message M’ related to the message Ml may further push to the users in the same group such as user terminal 10.4 and user terminal 10B. On the other hand, if tire user terminal 10C and user terminal IOC belongs to different groups, the message Ml and the related message M ^' may not be recommended to the user terminal 10C. in some embodiments, the comparing unit 140 may compare the real-time feedback FI, F2 and F3 on message M with the historical feedback FI, F2 and F3 from a plurality of users such as the user terminal 10A, IOB and IOC. If the real-time feedback F2 is similar to the historical feedback FI, it may be predicted that tire user terminal 10B has the preference P2 similar to the preference Pi of the user terminal 10A. Furthermore, If the real-time feedback F2. is similar to the historical feedback FI, the riser terminal 10B may be classified into the same group as the user terminal IOA. On the other hand, If the real-time feedback F3 is different from the historical feedback FI, the preference P3 of the user terminal IOC may ⁷ not be predicted according to the preference PI of the user terminal IOA. the user terminal IOC may be classified into different groups from the user terminal 10A.

In some embodiments, the comparing unit 140 may compare the preference PI, P2 and P3 on the message M from a plurality ⁷ of users such as the user terminal 10A, IOB and 10C. Even if the user terminal 10A, IOB and IOC have different feedback FI, F2 and F3, they may have the same preference PI, P2 and P3 for the message M. For example, the message M recommends the user terminal 10.A and the user terminal IOB a streamer, with respect to the message M, the user terminal 10A checks the streamer’s profile and the user terminal IOB watches the streamer. Even if their behaviors are not the same, both of their behaviors are positive. Therefore, the preference PI, P2 for the message M are similar, if the preference PI is similar to the preference P2, the user terminal 10A and the user terminal IOB may he classified into the same group. On the other hand, if the preference Pi is different from the preference P3, the user terminal IOA and the user terminal IOC may be classified into different groups. in some embodiments, Even if the user terminals 10A, IOB and IOC have similar feedback FI, F2 and F3, they may have different preferences PI, P2 and P3 tor the message M. For example, the message M recommends the user terminal iOA and the user terminal 1QB to watch a streamer, with respect to the message M, the user terminal 10A comments on the streamer with an angry emoji but the user terminal 10B comments with a smile emoji. Even if their behaviors are the same, their preference PI, P2 for the message M are different. Therefore, for the users in the same group, they may be further classified into subgroups. For example, if the user terminal IOA, J OB and IOC have a similar preference PI, P2 and P3 tor the message, user terminal 10A, 10B and IOC may ^¬ be classified into the same group. With respect to the similar preference PI, P2 and P3, if the feedback FI and the feedback F2 are similar, the user terminal IOA and the user terminal 10B may further be classified into the same subgroup. The users in the subgroup will have much similarity than the users not in the subgroup.

In some embodiments, if the message M is related to a streamer, the pushing unit 130 may further push the feedback F on the message M from a plurality of user terminal 10 to the streamer. The streamer may take the feedback F as a reference to improve his or her performance. For example, the message M may recommend the viewers to check the streamer’s singing. However, most of the viewers do not click the message M. It shows that most of the viewers are not interested in the streamer’s singing. The pushing unit 130 may push the feedback F and the preference P to the streamer and the streamer may understand that the viewers are not interested in his or her singing. The streamer may perform a dance next time instead of singing. In some embodiments, the message M may include a report and suggestions to the streamer. The report may include the monthly report of the streamer such as the streamer’s received coins, streaming hours, or streaming valid hours.

Furthermore, the system and method described in the above embodiments may be provided with a computer-readable non-transitory storage device such as a solid-state memory- device, an optical disk storage device, or a magnetic disk storage device. Alternatively, the programs may be downloaded from a server via the Internet.

Although technical content and features of the present invention are described above, a person having common knowledge in the technical field of the present invention may still make many variations and modifications without disobeying the teaching and disclosure of the present invention. Therefore, the scope of the present invention is not limited to the embodiments that are already disclosed, but includes another variation and modification that do not disobey the present invention, and is the scope covered by the following patent application scope.

Description of reference numerals 1 Communication system

10 User terminal

20 Server

100 Message distribution system

M Message

Mu U ser message

F Feedback

Freq Frequency

P Preference

110 Collecting unit

120 Evaluating unit

121 Message analysis element

122 Frequency analysis element

123 Relevance analysis element

124 preference evaluating element

1241 criteria calculator

1242 criteria evaluator

1243 criteria module

1244 weighting module

1245 training pairing element

1246 model generator

130 Pushing unit

140 Comparing Unit