Live broadcast interactive recommendation system based on real-time data analysis and feedback

Abstract

The invention relates to the technical field of big data live broadcast, in particular to a live broadcast interactive recommendation system based on real-time data analysis and feedback, which comprises an article preference analysis module, a price preference analysis module, a live broadcast auxiliary module and a feedback optimization module. According to the method, the preferred item represents the item type expected to be purchased by the user at present, the relative purchase coefficient represents the price proportion acceptable by the user, iteration updating can be continuously carried out along with feedback of each time of live broadcast, and the two kinds of data are analyzed in advance, so that the user does not need to purchase the item when entering the live broadcast room. The system can directly establish live broadcast interaction auxiliary elements for assisting anchor interaction without waiting for users to perform related operations, thereby greatly improving the flexibility and real-time performance of data analysis during live broadcast, and solving the problem that in the prior art, the live broadcast interaction auxiliary elements cannot be directly established under the condition that the behavior information of live broadcast audiences is less. And accurate user analysis cannot be carried out to assist live broadcast.

Description

Technical Field

[0001] The present invention relates to the field of big data live broadcast technology, and in particular to a live broadcast interactive recommendation system based on real-time data analysis and feedback. Background Art

[0002] Live streaming is a new e-commerce model that uses live streaming technology on the Internet platform to display products, answer inquiries and provide shopping guide services. This model not only enhances interactivity and improves affinity, but also achieves price advantages by directly connecting products and consumers. During the live broadcast process, the interaction between the anchor and the audience is crucial, which can enhance the audience's participation and activity, thereby increasing the attractiveness of the live broadcast and the sales conversion rate.

[0003] However, with the surge in the number of users watching live broadcasts, it is becoming increasingly difficult to understand and respond to user needs in real time. Especially for popular anchors, it is difficult to quickly and accurately capture users' specific needs and feedback in the face of massive and quickly flashing bullet screens and comments. Although existing technologies can assist live broadcasts by analyzing user behavior in the live broadcast room, these technologies still have some defects, especially in the cold start problem.

[0004] The cold start problem is mainly reflected in two aspects: first, for new users, especially those with less behavior, it is difficult for even popular anchors to conduct effective analysis and interaction; second, for small anchors, due to the small number of viewers, analysis and assistance are more difficult. Therefore, people need a live interactive recommendation system that can still perform accurate user analysis to assist live broadcasts when there is less information about live audience behavior. Summary of the invention

[0005] Therefore, the present invention provides a live interactive recommendation system based on real-time data analysis and feedback, so as to solve the problem in the prior art that accurate user analysis cannot be performed to assist the live interactive recommendation system in the live broadcast when there is little information on live audience behavior.

[0006] The present invention provides a live interactive recommendation system based on real-time data analysis and feedback, comprising:

[0007] An item preference analysis module is used to obtain the real-time browsing history of each user and obtain the preferred items according to the real-time browsing history of each user, wherein the preferred items represent the types of items that the user currently expects to purchase;

[0008] A price preference analysis module is used to obtain the relative purchase coefficient of each user's preferred item, wherein the relative purchase coefficient represents the ratio of the price that the user can accept to purchase the preferred item to the average transaction price of the preferred item;

[0009] A live broadcast auxiliary module is used to obtain the distribution of the preferred items of users in the target live broadcast room, and obtain live broadcast interactive auxiliary elements according to the distribution of the preferred items and the relative purchase coefficient of the preferred items corresponding to each user in the target live broadcast room, wherein the live broadcast interactive auxiliary elements are used to assist the anchor to interact with the distribution of the preferred items of users in the target live broadcast room;

[0010] The feedback optimization module is used to correct the relative purchase coefficient of each user's preferred items based on the user's transaction status based on the live interactive auxiliary elements as feedback.

[0011] In a preferred embodiment: the live broadcast interactive auxiliary elements include the anchor prompt element and the viewing interactive element; the distribution of the preferred items of the users in the target live broadcast room is obtained, and the live broadcast interactive auxiliary elements are obtained according to the distribution of the preferred items and the relative purchase coefficient of the preferred items corresponding to each user in the target live broadcast room, including:

[0012] Count the distribution of users' preferred items in the target live broadcast room and obtain the main items and secondary items;

[0013] Establish anchor prompt elements based on the relative purchase coefficient of each user in the target live broadcast room corresponding to the main items;

[0014] The live segments of the secondary items in the live broadcast record are intercepted, and interactive elements for viewing the live segments of each secondary item are established.

[0015] In a preferred embodiment, the transaction status of the user based on the live interactive auxiliary elements is used as feedback to correct the relative purchase coefficient of each user's preferred items, including:

[0016] Using the transaction status of users based on the anchor's prompt elements as feedback, correct the relative purchase coefficient of the main item for each user who prefers the main item in the live broadcast room;

[0017] Based on the users' viewing volume and transaction status of the interactive elements, the relative purchase coefficient of secondary items for each user in the live broadcast room who prefers secondary items is corrected.

[0018] In a preferred embodiment: based on the viewing amount and transaction status of the user viewing the interactive element as feedback, the relative purchase coefficient of each user who prefers the secondary item in the live broadcast room is corrected for the secondary item, including:

[0019] The relative purchase coefficient of each user who prefers secondary items in the live broadcast room is corrected by the following formula:

[0020] ;

[0021] in, Represents a user's relative purchase coefficient for a secondary item after correction. Indicates the user's relative purchase coefficient for the secondary item before correction. Indicates the number of views of the interactive element corresponding to the secondary item. A Boolean value indicating whether the user purchased the secondary item. is 1, otherwise it is 0. Indicates the ratio of the actual selling price of the secondary item to the average transaction price of the secondary item. The first preset linear function, and They represent different preset weight coefficients respectively.

[0022] In a preferred embodiment: according to the relative purchase coefficient of each user in the target live broadcast room corresponding to the main item, the anchor prompt element is established, including:

[0023] Minimum cost to acquire key items;

[0024] Calculate the mean of the relative purchase coefficients of each user in the target live broadcast room corresponding to the main items to obtain the average purchase coefficient;

[0025] Obtain the target expected selling price based on the minimum cost and the average purchase coefficient;

[0026] Create anchor prompt elements based on target desired selling price.

[0027] In a preferred embodiment: the preferred items also include empty items, and the empty items are used to indicate that the type of item that the user currently wants to purchase cannot be determined temporarily; the distribution of the preferred items of the users in the target live broadcast room is statistically analyzed to obtain the main items and the secondary items, including:

[0028] If the proportion of users who prefer empty items in the target live broadcast room is the largest, extract the user's evaluation information;

[0029] According to the user's evaluation information, the main items and the secondary items are obtained.

[0030] In a preferred embodiment, obtaining the main items and the secondary items according to the user's evaluation information includes:

[0031] Extract keywords from the evaluation information and obtain the evaluation word vector of each keyword based on the preset word embedding model;

[0032] Get the product word vectors of the items sold in the target live broadcast room;

[0033] The selling items corresponding to the product word vectors closest to multiple evaluation word vectors are selected as the main items, and other selling items in the target live broadcast room are selected as secondary items.

[0034] In a preferred embodiment: obtaining the relative purchase coefficient of each user's preferred items includes:

[0035] Get a user's real-time browsing history for a preferred item;

[0036] Obtain the user's purchase record for the preferred item;

[0037] Get the average purchase price of the preferred item;

[0038] According to the real-time browsing record price, purchase record price and average purchase price of the preferred item, the relative purchase coefficient of the user for the preferred item is obtained.

[0039] In a preferred embodiment, obtaining the user's relative purchase coefficient for the preferred item based on the real-time browsing record price, purchase record price, and average purchase price of the preferred item includes:

[0040] The relative purchase coefficient of a user for a preferred item is calculated by the following formula:

[0041] ;

[0042] in, Represents a user's relative purchase coefficient for a preferred item, Indicates the average price of the user's real-time browsing history for the preferred item. Indicates the average price of the user's purchase records for the preferred item. represents the average purchase price of the preferred item, and Respectively represent different preset weight parameters, Represents the second preset linear function.

[0043] In a preferred embodiment: obtaining the relative purchase coefficient of each user's preferred items includes:

[0044] Obtain similar items to a user's preferred items, as well as the relative purchase coefficient of similar items corresponding to the user;

[0045] According to the relative purchase coefficient of the similar items corresponding to the user, similar items of the preferred items are obtained.

[0046] The beneficial effects of adopting the above embodiment are:

[0047] The present invention provides a live broadcast interactive recommendation system based on real-time data analysis and feedback, comprising an item preference analysis module, a price preference analysis module, a live broadcast auxiliary module and a feedback optimization module, which first obtains the real-time browsing record of each user, obtains the preferred items according to the real-time browsing record of each user, then obtains the relative purchase coefficient of the preferred items of each user, and then obtains the distribution of the preferred items of the users in a target live broadcast room, and obtains the live broadcast interactive auxiliary elements according to the distribution of the preferred items and the relative purchase coefficient of each user in the target live broadcast room corresponding to the preferred items, and finally uses the transaction status of the users based on the live broadcast interactive auxiliary elements as feedback to correct the relative purchase coefficient of the preferred items of each user. Compared with the prior art, the preferred items in the present invention represent the types of items that the user currently expects to buy, and the relative purchase coefficient represents the ratio of the price that the user can accept to the average transaction price of the preferred items, and can be continuously iterated and updated with the feedback of each live broadcast. By analyzing the above two types of data in advance, when the user enters the live broadcast room, the system can directly establish live broadcast interactive auxiliary elements for assisting the host interaction through the above two types of data without waiting for the user to perform relevant operations, which greatly improves the flexibility and real-time performance of data analysis during live broadcast, and solves the problem in the prior art that the live broadcast interactive recommendation system cannot perform accurate user analysis to assist the live broadcast when there is less information on the behavior of the live broadcast audience. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A system architecture diagram of an embodiment of a live interactive recommendation system based on real-time data analysis and feedback provided by the present invention;

[0049] Figure 2 A diagram of the method steps performed by the live interactive recommendation system based on real-time data analysis and feedback provided by the present invention;

[0050] Figure 3 for Figure 2 Specific step diagram of step S203 in FIG. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Combination Figure 1 and 2As shown, a specific embodiment of the present invention discloses a live interactive recommendation system based on real-time data analysis and feedback, including an item preference analysis module 110, a price preference analysis module 120, a live auxiliary module 130 and a feedback optimization module 140, and the above four modules are respectively used to perform the following four steps:

[0053] S201, obtaining the real-time browsing record of each user, and obtaining the preferred items according to the real-time browsing record of each user, wherein the preferred items represent the types of items that the user currently expects to purchase;

[0054] S202, obtaining a relative purchase coefficient of each user's preferred item, wherein the relative purchase coefficient represents a ratio of a price that the user can accept to purchase the preferred item to an average transaction price of the preferred item;

[0055] S203, obtaining the distribution of the preferred items of the users in the target live broadcast room, and obtaining a live broadcast interaction auxiliary element according to the distribution of the preferred items and the relative purchase coefficient of the preferred items corresponding to each user in the target live broadcast room, wherein the live broadcast interaction auxiliary element is used to assist the anchor to interact with the distribution of the preferred items of the users in the target live broadcast room;

[0056] S204: Using the transaction status of the users based on the live interactive auxiliary elements as feedback, the relative purchase coefficient of each user's preferred items is corrected.

[0057] The four modules mentioned above all need to analyze massive amounts of data, so they can be deployed in a server independent of the live broadcast end or the user end to avoid affecting the user experience.

[0058] Real-time browsing records refer to the browsing records of users for products in a recent set period of time, which to a certain extent reflect the products that the users are interested in. The time for counting real-time browsing records can be flexibly defined according to the specific situation, such as one month or one week before the current moment, or the time period from the online time to the current moment. Similarly, the execution time of step S201 is generally also set according to this defined time period, such as regularly executing once a month or automatically running in the background every time the user goes online for a period of time. It only needs to ensure that it has been executed at least once before the user enters the live broadcast room. The execution time of step S202 is the same.

[0059] According to the real-time browsing records, the user's preferred items can be analyzed, and the preferred items indicate the types of items that the user currently expects to buy. In practice, the preferred items can also be flexibly determined according to actual needs. The preferred items can be specific to the brand and model of a certain kind of goods, or they can be a broader category of items, such as T-shirts, mobile phones, bread, etc. In practice, the second method is preferred. The definition of the type of item can be set by itself, or it can be set using the existing word embedding technology. Specifically, in word embedding technology, each word corresponds to a word vector, and the word vectors of words with similar meanings are similar, while the word vectors of words with different meanings are quite different. For example, the word vectors of a water cup and a kettle are relatively similar, while the word vectors of a water cup and a car are more different. In practice, the specific types of preferred items can be established directly based on the vocabulary defined by the word embedding technology. In addition, it can be understood that there can be multiple preferred items corresponding to a user.

[0060] The relative purchase coefficient in the above content indicates the ratio of the price of the preferred item that the user can accept to the average transaction price of the target preferred item. It can be understood that the relative purchase coefficient can be obtained based on the user's purchase record. The relative purchase coefficient in the present invention is a fuzzy approximate representation and does not need to be too precise. When calculating the relative purchase coefficient, the price of the preferred item that the user can accept can be expressed by obtaining the lowest price or the highest price at which the user purchases the item. The average transaction price is the average price at which all users purchase the item. In the case of a low number of samples, the average transaction price can also use more authoritative values ​​such as the official guide price.

[0061] Live interactive auxiliary elements are virtual elements used to assist the host in live broadcasting. Depending on specific needs, they can be text dialog boxes, keywords, prompt sounds, charts and other elements (with certain prompt content) to prompt the host, or they can be links, buttons, etc. that can interact with users.

[0062] Compared with the prior art, the preferred items in the present invention represent the types of items that the user currently expects to buy, and the relative purchase coefficient represents the ratio of the price that the user can accept to the average transaction price of the preferred items, and can be continuously iterated and updated with the feedback of each live broadcast. By analyzing the above two types of data in advance, when the user enters the live broadcast room, the system can directly establish live broadcast interactive auxiliary elements for assisting the host interaction through the above two types of data without waiting for the user to perform relevant operations, which greatly improves the flexibility and real-time performance of data analysis during live broadcast, and solves the problem in the prior art that the live broadcast interactive recommendation system cannot perform accurate user analysis to assist the live broadcast when there is less information on the behavior of the live broadcast audience.

[0063] Furthermore, in the foregoing description, it is assumed that each user has an initial relative purchase coefficient for each preferred item. Initially, each relative purchase coefficient may be a preset value. The present invention further provides a preferred embodiment. In this embodiment, the above step S202, obtaining the relative purchase coefficient of each user's preferred item, specifically includes:

[0064] Get a user's real-time browsing history for a preferred item;

[0065] Obtain the user's purchase record for the preferred item;

[0066] Get the average purchase price of the preferred item;

[0067] According to the real-time browsing record price, purchase record price and average purchase price of the preferred item, the relative purchase coefficient of the user for the preferred item is obtained.

[0068] The above process is mainly aimed at the situation where there is no preset relative purchase coefficient, but the user has a browsing or purchase record of the preferred item. This embodiment mainly measures the user's price preference for a preferred item through three dimensions: the price of the real-time browsing record, the price of the purchase record, and the average purchase price. Among them, the price of the real-time browsing record represents the price that the user is referring to at this time; the price of the purchase record represents a clear price that the user can accept; the average purchase price is the price at which all users purchase this preferred item, which represents a universal price and serves as a standard for measuring the price acceptable to users. Through the joint analysis of the above three dimensions, a more scientific and reasonable initial relative purchase coefficient can be obtained than the preset value, thereby improving the accuracy of subsequent analysis.

[0069] Specifically, in a preferred embodiment, the above step of obtaining the user's relative purchase coefficient for the preferred item based on the real-time browsing record price, purchase record price, and average purchase price of the preferred item specifically includes:

[0070] The relative purchase coefficient of a user for a preferred item is calculated by the following formula:

[0071] ;

[0072] in, Represents a user's relative purchase coefficient for a preferred item, Indicates the average price of the user's real-time browsing history for the preferred item. Indicates the average price of the user's purchase records for the preferred item. represents the average purchase price of the preferred item, and Respectively represent different preset weight parameters, Represents the second preset linear function.

[0073] It should be noted that, in the above process, the preset weight parameter and the second preset linear function are both artificially preset values ​​that can be obtained based on experiments or experience.

[0074] Furthermore, in practice, most items do not meet the conditions for users to purchase frequently. Therefore, in practice, it is very likely that a preferred item not only has no preset relative purchase coefficient, but also has no sufficient browsing or purchase records that can be analyzed. Therefore, in response to the above situation, the present invention also provides a more preferred embodiment to obtain a more reasonable relative purchase coefficient when the amount of relevant data of the preferred item is small.

[0075] Specifically, in a preferred embodiment, the above step S202, obtaining the relative purchase coefficient of each user's preferred items, further includes:

[0076] Obtain similar items to a user's preferred items, as well as the relative purchase coefficient of similar items corresponding to the user;

[0077] According to the relative purchase coefficient of the similar items corresponding to the user, similar items of the preferred items are obtained.

[0078] Different from the previous embodiment, the above process is mainly applicable to the case where there is no preset relative purchase coefficient, that is, there are few browsing and purchase records related to the preferred items. This embodiment analyzes the relative purchase coefficients of existing items similar to the preferred items to obtain coefficients that conform to the user's behavior habits. For example, if the preferred item is a tablet computer, and a user purchases a tablet computer for the first time, there are few records related to the tablet computer in the system. Then, the relative purchase coefficients of similar items to the tablet computer (such as mobile phones, laptops, computer accessories, etc.) can be obtained, and the average of these coefficients (which can also be weighted and summed according to the degree of similarity) is taken as the initial relative purchase parameter of the tablet computer. When the user's relative purchase coefficients for mobile phones, laptops, computer accessories, etc. are all high, it means that the user can accept a price higher than the market price in the field of electronic products. At this time, the relative purchase coefficient corresponding to the tablet computer is naturally high, and vice versa.

[0079] Similarly, how to judge whether an item is similar to a preferred item can be manually defined or directly implemented by the word embedding technology mentioned above. The specific process can be:

[0080] Create the word vector of the preferred item as the first word vector;

[0081] Get the word vector of each preset candidate item as the second word vector;

[0082] Calculate the similarity between each second word vector and the first word vector (for example, calculate the Euclidean distance, etc.), and then select the items corresponding to the first several second word vectors with the highest similarity as similar items.

[0083] Further, combined with Figure 3 As shown, in a preferred embodiment, the live broadcast interactive auxiliary elements include the anchor prompt element and the viewing interactive element; the above step S203, obtaining the distribution of the preferred items of the users in the target live broadcast room, and obtaining the live broadcast interactive auxiliary elements according to the distribution of the preferred items and the relative purchase coefficient of the preferred items corresponding to each user in the target live broadcast room, specifically includes:

[0084] S301, counting the distribution of users' preferred items in the target live broadcast room to obtain main items and secondary items;

[0085] S302, establishing a host prompt element according to the relative purchase coefficient of each user in the target live broadcast room corresponding to the main item;

[0086] S303: intercept the live broadcast segments of the secondary items in the live broadcast record, and establish a viewing interactive element for each live broadcast segment of the secondary items.

[0087] In the above process, the main items are the items that most viewers in the live broadcast room expect to buy, that is, the preferred items that account for the largest proportion of users in the live broadcast room, and the secondary items are the non-main items sold in the live broadcast room. Obviously, both the main items and the secondary items must be items that can be sold in the live broadcast room. Similar to the preferred items, there is also a one-to-one correspondence between the main items and the secondary items and the users. The following text will assume this one-to-one correspondence without explaining it in detail everywhere. For example, the relative purchase coefficient of the main item is the relative purchase coefficient of a user corresponding to the preferred item indicated by the main item.

[0088] The above process establishes anchor prompt elements and viewing interactive elements for customers corresponding to the main items and secondary items, respectively. In this way, this embodiment, on the one hand, takes care of the needs of most viewers in the live broadcast room, and reminds the anchor and the audience to interact through anchor prompt elements (such as a distribution pie chart of user preferences, a user purchasing power portrait formed according to the user's relative purchase coefficient, etc.), so as to meet the needs of the audience corresponding to the main items. On the other hand, it also meets the needs of other viewers of the secondary items in the live broadcast room by viewing interactive elements. The audience corresponding to the secondary items can watch the live broadcast clip of the anchor explaining the secondary items by viewing interactive elements (such as links, buttons, etc.) (the clip can be preset by humans, or can be intercepted by AI from past recordings or content before this live broadcast). Based on the preferred items and relative purchase coefficients analyzed in the previous article, this embodiment further improves the user's interactive experience by dividing the anchor prompt elements and viewing interactive elements. On the premise of further reducing the information that users need to input (such as sending barrages to ask the anchor, etc.), all users' expectations can be responded to in a timely manner, further improving the flexibility and real-time performance of the present invention.

[0089] Similar to the determination method of the relative purchase coefficient in the previous text, the content in the previous text assumes that each user can analyze and obtain a preferred item. In practice, due to the small number of user usage records, it is also possible that an exact preferred item cannot be analyzed. Therefore, in a preferred embodiment, the definition of preferred items also includes empty items. Empty items are used to indicate that the type of item that the user currently expects to purchase cannot be determined temporarily. On this basis, the above step S301, statistics on the distribution of the preferred items of users in the target live broadcast room, obtains the main items and secondary items, specifically including:

[0090] If the proportion of users who prefer empty items in the target live broadcast room is the largest, extract the user's evaluation information;

[0091] According to the user's evaluation information, the main items and the secondary items are obtained.

[0092] In the above content, evaluation information refers to chat information such as bullet screens and comments sent by users when watching live broadcasts. This embodiment is mainly aimed at the situation where user information is further missing, that is, the user information is so little that the user's preferred items cannot be determined. At this time, the system starts with the user's evaluation information and associates it with the user's evaluation information to obtain items that the user may be interested in as the main items (other items sold in the live broadcast room are naturally secondary items), thereby further improving the interactivity of the live broadcast under the condition of missing information.

[0093] Specifically, in a preferred embodiment, the above step of obtaining the main items and the secondary items according to the user's evaluation information specifically includes:

[0094] Extract keywords from the evaluation information and obtain the evaluation word vector of each keyword based on the preset word embedding model;

[0095] Get the product word vectors of the items sold in the target live broadcast room;

[0096] The selling items corresponding to the product word vectors closest to multiple evaluation word vectors are selected as the main items, and other selling items in the target live broadcast room are selected as secondary items.

[0097] The significance of the above process is that the word embedding technology is used to guess the items that the user is interested in. For example, based on the words such as breathability, fit, size, and size in the user evaluation information, it can be guessed that the main items that the user is interested in may be clothing, and based on the words such as smoothness, battery life, memory, and chip signal in the user evaluation information, it can be guessed that the main items that the user is interested in may be mobile phones. The preset word embedding model in the above process is a pre-set model that converts words and word vectors into each other that meets the application scenario of the present invention.

[0098] The word embedding technology itself is a technology in the field of natural language recognition, and its conventional purpose is to guess the probability of the next word. For example, the existing CHATGPT technology can be simply explained as predicting the probability of the next word in a sentence through word embedding technology. In this embodiment, the technology is used to guess the associative items of interest to the user to improve the applicability of the system. It can be understood that the principle of word embedding technology has been briefly described above, and it is itself an existing technology that can be understood by those skilled in the art, so this article will not explain it in detail.

[0099] Furthermore, in a preferred embodiment, the above step S302, based on the relative purchase coefficient of each user in the target live broadcast room corresponding to the main item, establishes the anchor prompt element, which specifically includes:

[0100] Minimum cost to acquire key items;

[0101] Calculate the mean of the relative purchase coefficients of each user in the target live broadcast room corresponding to the main items to obtain the average purchase coefficient;

[0102] Obtain the target expected selling price based on the minimum cost and the average purchase coefficient;

[0103] Create anchor prompt elements based on target desired selling price.

[0104] The significance of the above process is to further use the definition of the relative purchase coefficient itself to calculate the price that meets the expectations of the current live broadcast room users, so as to serve as a host prompt element to assist the host in interacting and increasing sales. It can be understood that the above method is only a preferred implementation method. In practice, the host prompt element can also be established in other forms, such as generating a distribution pie chart of the relative purchase coefficient of the live broadcast room users for a certain item, so that the host can timely understand the purchasing power of the live broadcast room users (especially for users who have just entered the live broadcast room or do not like to speak), so as to adjust strategies such as live broadcast copywriting.

[0105] Furthermore, in a preferred embodiment, the above step S204, using the transaction status of the user based on the live interactive auxiliary element as feedback, corrects the relative purchase coefficient of each user's preferred items, specifically including:

[0106] Using the transaction status of users based on the anchor's prompt elements as feedback, correct the relative purchase coefficient of the main item for each user who prefers the main item in the live broadcast room;

[0107] Based on the users' viewing volume and transaction status of the interactive elements, the relative purchase coefficient of secondary items for each user in the live broadcast room who prefers secondary items is corrected.

[0108] This embodiment can be understood as another advantage of the design of dividing the live interactive auxiliary elements into anchor prompt elements and viewing interactive elements in the previous text, that is, this embodiment allows the system to simultaneously correct the relative purchase coefficients of major items and minor items in a live broadcast to improve the accuracy of the system. As the number of live broadcasts increases, this embodiment can significantly increase the speed at which the relative purchase coefficients saved in the system converge to accurate data.

[0109] It is understandable that the transaction status of the user based on the anchor prompt element in the above process can be understood as the transaction status of a preferred item after the anchor responds based on the anchor prompt element, and the specific method of correcting the relative purchase coefficient according to this situation can also be set according to the actual situation. After the anchor responds based on the anchor prompt element (for example, raising the price of the product), if there is a user transaction, then the relative purchase coefficient of the transaction user for this type of product can be increased. For another example, after the anchor responds based on the anchor prompt element (for example, the anchor performs a program that suits the audience's taste), if the user transaction rate increases, then the relative purchase coefficient of the transaction user for this type of product can be adjusted according to the transaction price.

[0110] Furthermore, compared with the above-mentioned method of adjusting the relative purchase coefficient for the main items, the same applies to the user's viewing volume and transaction status based on the interactive elements. However, due to the different specific types of live interactive auxiliary elements, the viewing volume can be further considered when adjusting the relative purchase coefficient of the secondary items.

[0111] Specifically, in a preferred embodiment, the above step: taking the viewing amount and transaction status of the interactive elements as feedback, correcting the relative purchase coefficient of each user who prefers the secondary item in the live broadcast room, specifically includes:

[0112] The relative purchase coefficient of each user who prefers secondary items in the live broadcast room is corrected by the following formula:

[0113] ;

[0114] in, Represents a user's relative purchase coefficient for a secondary item after correction. Indicates the user's relative purchase coefficient for the secondary item before correction. Indicates the number of views of the interactive element corresponding to the secondary item. A Boolean value indicating whether the user purchased the secondary item. is 1, otherwise it is 0. Indicates the ratio of the actual selling price of the secondary item to the average transaction price of the secondary item. The first preset linear function, and They represent different preset weight coefficients respectively.

[0115] The significance of the above formula is that it takes into account both the viewing volume and the transaction situation, and corrects the relative purchase coefficient of the secondary items. On the one hand, the higher the viewing volume, the more attractive the price of the secondary items in the corresponding live broadcast segment is to the user. At this time, appropriate adjustments can be made according to the level. On the other hand, the user's transaction behavior directly indicates the price acceptable to the user. At this time, corrections can be made based on the difference between the ratio corresponding to the actual transaction price and the relative purchase coefficient. Similarly, the preset weight coefficient and the first preset linear function in the above formula are also preset parameters that can be obtained based on experiments or experience.

[0116] The present invention provides a live broadcast interactive recommendation system based on real-time data analysis and feedback, comprising an item preference analysis module, a price preference analysis module, a live broadcast auxiliary module and a feedback optimization module, which first obtains the real-time browsing record of each user, obtains the preferred items according to the real-time browsing record of each user, then obtains the relative purchase coefficient of the preferred items of each user, and then obtains the distribution of the preferred items of the users in a target live broadcast room, and obtains the live broadcast interactive auxiliary elements according to the distribution of the preferred items and the relative purchase coefficient of each user in the target live broadcast room corresponding to the preferred items, and finally uses the transaction status of the users based on the live broadcast interactive auxiliary elements as feedback to correct the relative purchase coefficient of the preferred items of each user. Compared with the prior art, the preferred items in the present invention represent the types of items that the user currently expects to buy, and the relative purchase coefficient represents the ratio of the price that the user can accept to the average transaction price of the preferred items, and can be continuously iterated and updated with the feedback of each live broadcast. By analyzing the above two types of data in advance, when the user enters the live broadcast room, the system can directly establish live broadcast interactive auxiliary elements for assisting the host interaction through the above two types of data without waiting for the user to perform relevant operations, which greatly improves the flexibility and real-time performance of data analysis during live broadcast, and solves the problem in the prior art that the live broadcast interactive recommendation system cannot perform accurate user analysis to assist the live broadcast when there is less information on the behavior of the live broadcast audience.

[0117] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0118] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A live interactive recommendation system based on real-time data analysis and feedback, characterized in that: include: An item preference analysis module is used to obtain the real-time browsing history of each user and obtain the preferred items according to the real-time browsing history of each user, wherein the preferred items represent the types of items that the user currently expects to purchase; A price preference analysis module is used to obtain the relative purchase coefficient of each user's preferred item, wherein the relative purchase coefficient represents the ratio of the price that the user can accept to purchase the preferred item to the average transaction price of the preferred item; A live broadcast auxiliary module is used to obtain the distribution of the preferred items of users in the target live broadcast room, and obtain live broadcast interactive auxiliary elements according to the distribution of the preferred items and the relative purchase coefficient of the preferred items corresponding to each user in the target live broadcast room, wherein the live broadcast interactive auxiliary elements are used to assist the anchor to interact with the distribution of the preferred items of users in the target live broadcast room; The feedback optimization module is used to correct the relative purchase coefficient of each user's preferred items based on the user's transaction status based on the live interactive auxiliary elements as feedback.

2. The live interactive recommendation system based on real-time data analysis and feedback according to claim 1, characterized in that: Live interactive auxiliary elements include anchor prompt elements and viewing interactive elements; The distribution of the preferred items of the users in the target live broadcast room is obtained, and according to the distribution of the preferred items and the relative purchase coefficient of the preferred items corresponding to each user in the target live broadcast room, the live broadcast interaction auxiliary elements are obtained, including: Count the distribution of users' preferred items in the target live broadcast room and obtain the main items and secondary items; Establish anchor prompt elements based on the relative purchase coefficient of each user in the target live broadcast room corresponding to the main items; The live segments of the secondary items in the live broadcast record are intercepted, and interactive elements for viewing the live segments of each secondary item are established.

3. The live interactive recommendation system based on real-time data analysis and feedback according to claim 2 is characterized in that: Based on the transaction status of users based on the interactive auxiliary elements of the live broadcast, the relative purchase coefficient of each user's preferred items is corrected, including: Using the transaction status of users based on the anchor's prompt elements as feedback, correct the relative purchase coefficient of the main item for each user who prefers the main item in the live broadcast room; Based on the users' viewing volume and transaction status of the interactive elements, the relative purchase coefficient of secondary items for each user in the live broadcast room who prefers secondary items is corrected.

4. The live interactive recommendation system based on real-time data analysis and feedback according to claim 3 is characterized in that: Based on the user's viewing volume and transaction status of the interactive elements, the relative purchase coefficient of each user who prefers secondary items in the live broadcast room is corrected, including: The relative purchase coefficient of each user who prefers secondary items in the live broadcast room is corrected by the following formula: ； in, Represents a user's relative purchase coefficient for a secondary item after correction. Indicates the user's relative purchase coefficient for the secondary item before correction. Indicates the number of views of the interactive element corresponding to the secondary item. A Boolean value indicating whether the user purchased the secondary item. is 1, otherwise it is 0. Indicates the ratio of the actual selling price of the secondary item to the average transaction price of the secondary item. The first preset linear function, and They represent different preset weight coefficients respectively.

5. The live interactive recommendation system based on real-time data analysis and feedback according to claim 2 is characterized in that: According to the relative purchase coefficient of each user in the target live broadcast room corresponding to the main items, establish the anchor prompt elements, including: Minimum cost to acquire key items; Calculate the mean of the relative purchase coefficients of each user in the target live broadcast room corresponding to the main items to obtain the average purchase coefficient; Obtain the target expected selling price based on the minimum cost and the average purchase coefficient; Create anchor prompt elements based on target desired selling price.

6. The live interactive recommendation system based on real-time data analysis and feedback according to claim 2 is characterized in that: Preference items also include empty items, which are used to indicate that the type of item that the user currently wants to purchase cannot be determined for the time being; Count the distribution of users' preferred items in the target live broadcast room and get the main items and secondary items, including: If the proportion of users who prefer empty items in the target live broadcast room is the largest, extract the user's evaluation information; According to the user's evaluation information, the main items and the secondary items are obtained.

7. The live interactive recommendation system based on real-time data analysis and feedback according to claim 6 is characterized in that: According to the user's evaluation information, get the main items and secondary items, including: Extract keywords from the evaluation information and obtain the evaluation word vector of each keyword based on the preset word embedding model; Get the product word vectors of the items sold in the target live broadcast room; The selling items corresponding to the product word vectors closest to multiple evaluation word vectors are selected as the main items, and other selling items in the target live broadcast room are selected as secondary items.

8. The live interactive recommendation system based on real-time data analysis and feedback according to claim 1, characterized in that: Get the relative purchase coefficient of each user's preferred items, including: Get a user's real-time browsing history for a preferred item; Obtain the user's purchase record for the preferred item; Get the average purchase price of the preferred item; According to the real-time browsing record price, purchase record price and average purchase price of the preferred item, the relative purchase coefficient of the user for the preferred item is obtained.

9. The live interactive recommendation system based on real-time data analysis and feedback according to claim 8, characterized in that: According to the real-time browsing record price, purchase record price and average purchase price of the preferred item, the relative purchase coefficient of the user for the preferred item is obtained, including: The relative purchase coefficient of a user for a preferred item is calculated by the following formula: ； in, Represents a user's relative purchase coefficient for a preferred item, Indicates the average price of the user's real-time browsing history for the preferred item. Indicates the average price of the user's purchase records for the preferred item. represents the average purchase price of the preferred item, and Respectively represent different preset weight parameters, Represents the second preset linear function.

10. The live interactive recommendation system based on real-time data analysis and feedback according to claim 1, characterized in that: Get the relative purchase coefficient of each user's preferred items, including: Obtain similar items to a user's preferred items, as well as the relative purchase coefficient of similar items corresponding to the user; According to the relative purchase coefficient of the similar items corresponding to the user, similar items of the preferred items are obtained.

Images