Information recommendation method and system based on artificial intelligence

Abstract

The invention relates to the technical field of data processing, in particular to an artificial intelligence-based information recommendation method and system, and the method comprises the steps: obtaining the push information of a user in a preset historical time period, and obtaining the information cocoon house depth of the user according to the push information of the user; according to the reading duration of the user for each piece of push information, obtaining the satisfaction degree of the user for the push information; combining the information cocoon house depth and the satisfaction degree to obtain an information recommendation degree; obtaining a difference between two adjacent reading durations of the user for the push information of the same attribute, and obtaining an information acceptance degree of the user for the push information of each attribute; according to the information recommendation degree and the information acceptance degree, information recommendation is performed on the user, more comprehensive data information can be recommended to the user, pushing limitation is eliminated, pushing diversity of the data information is realized, and the user can effectively contact new interests or potential demands.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to an information recommendation method and system based on artificial intelligence. Background Art

[0002] The Internet has gradually become the main platform for people to obtain information. Most modern Internet platforms rely on complex algorithms to push personalized content based on users' interests, browsing records, and social behaviors. This recommendation mechanism seemingly improves the user experience and makes information acquisition more accurate and efficient. However, it also makes users more and more easily confined to their known fields during the browsing process. Users mostly see content they recognize or are interested in, while information that is contrary to this is difficult to come into view, resulting in an increase in the singularity and one-sidedness of information. This information "filter bubble" gradually forms the so-called "information cocoon room", and people's horizons are gradually "circumscribed" by algorithms, ignoring diverse viewpoints and comprehensive information. As a result, people's cognition and horizons become narrow, and the exposure to new and different thoughts and knowledge becomes less and less.

[0003] Existing information recommendation algorithms usually perform personalized recommendations by analyzing the similarity of users' interests and behaviors. Specifically, first, it is assumed that users prefer information with certain specific attributes, and then based on the interaction behaviors between users and this information, the understanding of users' true interests is gradually adjusted and optimized. Through data mining and analysis algorithms, the recommendation system can identify information with similar attributes and recommend it to users. However, this method has certain limitations: it usually only recommends information that is highly similar to users' historical interests, overly relying on personalization, which will lead to the lack of information diversity and limit the opportunities for users to contact new interests or potential needs. Summary of the Invention

[0004] In order to solve the technical problem that the information recommended by existing information recommendation algorithms for users has certain limitations, the purpose of the present invention is to provide an information recommendation method and system based on artificial intelligence, and the specific technical solutions adopted are as follows: In the first aspect of the present invention, an information recommendation method based on artificial intelligence is provided, including: Obtain the push information for the user within a preset historical time period, and obtain the depth of the user's information cocoon based on the push information for the user; Obtain the satisfaction degree of the user with the push information according to the viewing duration of each push information for the user within the preset historical time period; Combine the depth of the information cocoon and the satisfaction degree to obtain the information recommendation degree of the user for the push information; Obtain the difference in the viewing durations of two adjacent views of the push information for the same attribute by the user, and based on the overall situation of the difference, obtain the information acceptance degree of the user for the push information of each attribute; Recommend information to the user according to the information recommendation degree and the information acceptance degree.

[0005] In an exemplary embodiment, the information recommendation method based on artificial intelligence further includes: Obtain a set of push information attribute vectors corresponding to the push information for the user within a preset historical time period, where the set of push information attribute vectors includes the information attribute vectors of the push information for the user at each sampling moment within the preset historical time period; The obtaining the depth of the user's information cocoon based on the push information for the user includes: Obtain the similarity between the information attribute vectors at any two adjacent sampling moments, and obtain the similarity less than the preset similarity threshold to obtain the target similarity; Take the position between the two sampling moments corresponding to each target similarity as the segmentation position, and segment the set of push information attribute vectors to obtain a plurality of similarity segments; Based on each similarity segment, obtain the depth of the user's information cocoon.

[0006] In an exemplary embodiment, the obtaining the depth of the user's information cocoon based on each similarity segment includes: Obtain the mean value of the similarities included in each similarity segment, and obtain the number of the information attribute vectors in each similarity segment; According to the mean value of the similarities, the number of the information attribute vectors, and the number of the similarity segments, obtain the information cocoon depth sub-features of each similarity segment, where the information cocoon depth sub-features are proportional to the mean value of the similarities and the number of the information attribute vectors, and inversely proportional to the number of the similarity segments; Obtain the sum value of the information cocoon depth sub-features of each similarity segment and normalize it to obtain the depth of the user's information cocoon.

[0007] In an exemplary embodiment, obtaining the satisfaction degree of the user for the push information according to the viewing duration of the user for each push information within a preset historical time period includes: Obtain the viewing duration of the user for each push information, and obtain the average viewing duration of all users for the same push information; Based on the difference between the viewing duration of the user for each push information and the average viewing duration of the push information, obtain the anti-fooling parameter of the user for the push information; Based on the anti-fooling parameters of each push message for the user, the viewing duration of each push message for the user, and the average viewing duration, obtain the satisfaction degree of the user for the push message.

[0008] In an exemplary embodiment, the calculation formula of the anti-fooling parameter is as follows: ; Where, represents the anti-fooling parameter of the th user for the th push message, represents the viewing duration of the th user for the th push message; represents the average viewing duration of all users for the th push message; represents the natural constant; represents the pi; The calculation formula of the satisfaction degree is as follows: ; Where, represents the satisfaction degree of the th user, represents the number of push messages, represents the normalization function.

[0009] In an exemplary embodiment, combining the information cocoon depth and the satisfaction degree, obtain the information recommendation degree of the user for the push message, including: According to the information cocoon depth and the satisfaction degree, obtain the timely adaptability of the user for the push message, and the timely adaptability is inversely proportional to the information cocoon depth and directly proportional to the satisfaction degree; According to the timely adaptability, obtain the information recommendation degree, and the information recommendation degree is inversely proportional to the timely adaptability.

[0010] In an exemplary embodiment, the calculation formula of the timely adaptability is as follows: ; Where, represents the timely adaptability of the th user, represents the satisfaction degree of the th user, represents the information cocoon depth of the th user, represents the natural constant; The calculation formula of the information recommendation degree is: ; wherein, represents the information recommendation degree of the th user.

[0011] In an exemplary embodiment, obtaining the difference between the adjacent two viewing durations of the push information of the user for the same attribute, and obtaining the information acceptance degree of the push information of the user for each attribute according to the overall situation of the difference, includes: The calculation formula of the information acceptance degree is as follows: ; ; wherein, represents the information acceptance degree of the th user for the push information of the th attribute, represents the expected parameter, represents the th viewing of the push information of the th user for the th attribute, represents the total number of viewings of the push information of the th user for the th attribute, represents the viewing duration of the th viewing of the push information of the th user for the th attribute, represents the viewing duration of the th viewing of the push information of the th user for the th attribute.

[0012] In an exemplary embodiment, according to the information recommendation degree and the information acceptance degree, performing information recommendation for the user, includes: Obtaining the attributes of the push information with the information acceptance degree greater than the preset information acceptance degree threshold to obtain the target attributes; Multiplying the information acceptance degree of each target attribute by the information recommendation degree to obtain the distribution weight of the push information of each target attribute; Pushing the push information of each target attribute to the user according to the distribution weight of the push information of each target attribute.

[0013] In a second aspect of the present invention, there is provided an information recommendation system based on artificial intelligence, including: a memory and a processor; the memory is connected to the processor; the memory is used for storing program instructions; the processor is used for implementing the above-mentioned information recommendation method based on artificial intelligence when the program instructions are executed.

[0014] The present invention has the following beneficial effects: The information recommendation method based on artificial intelligence provided by the present invention analyzes the push information for the user within a preset historical time period, respectively obtains the depth of the user's information cocoon and the satisfaction degree with the push information, and then comprehensively obtains the information recommendation degree of the user for the push information based on these two aspects of data information. And according to the difference in the viewing duration of the user for the push information of the same attribute in two adjacent times, the information acceptance degree of the user for the push information of each attribute is obtained. Finally, according to the information recommendation degree and the information acceptance degree, information recommendation is performed for the user, which can recommend more comprehensive data information for the user, eliminate the push limitation, avoid only recommending information highly similar to the user's historical interests, realize the diversity of data information push, and enable the user to effectively contact new interests or potential needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flowchart of an information recommendation method based on artificial intelligence provided by an embodiment of the present invention; Figure 2 is a flowchart for obtaining the depth of the information cocoon provided by an embodiment of the present invention; Figure 3 is a flowchart of steps 1-3 provided by an embodiment of the present invention; Figure 4 is a logical relationship diagram between the curve of the depth of the information cocoon provided by an embodiment of the present invention and other related curves; Figure 5 is a flowchart for obtaining the satisfaction degree provided by an embodiment of the present invention; Figure 6 is a logical relationship diagram between the viewing speed of the user for the push information and the satisfaction degree provided by an embodiment of the present invention; Figure 7 is a flowchart for obtaining the information recommendation degree provided by an embodiment of the present invention; Figure 8 is a flowchart for performing information recommendation for the user provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following describes in detail the specific implementation manners, structures, features and effects of the present invention in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this invention belongs.

[0018] When a user receives information, generally, the information push platform analyzes the user's browsing behavior of the pushed information. By utilizing the similarity of the user's acceptance behavior of the pushed information with different attributes, a user portrait is created to identify the pushed information that the user is interested in, and then an information push mechanism is used to push the information that the user is interested in for each user. Although this push behavior can allow the user to continuously receive the information they are interested in, most of these interested information belongs to information with the same attribute or highly similar attributes. When the user obtains the pushed information with the same attribute for a long time, "aesthetic fatigue" in information acquisition will occur due to the excessive abundance of information with the same attribute, and the information cocoon will be generated because there is no collision of ideas with other attributes. This is an extremely unfavorable phenomenon for the sustainable operation of the platform. Therefore, the present invention proposes an information recommendation method and system based on artificial intelligence, which analyzes the user's acceptance behavior of the pushed information and jointly pushes information with other attributes when the user has a high degree of intolerance for the pushed information with the current attribute.

[0019] As Figure 1 shown, an information recommendation method based on artificial intelligence provided in this embodiment includes the following steps: Step 1: Obtain the pushed information for the user within a preset historical time period, and obtain the depth of the user's information cocoon according to the pushed information for the user.

[0020] The main purpose to be achieved by the present invention is to push information across domains (i.e., attributes) for the user, and its main underlying logic is to analyze the similarity of the pushed information received by the user and the user's information acceptance behavior. Therefore, it is necessary to collect the information pushed by the platform for the user and the user's information reception behavior.

[0021] Taking any platform as an example, a user library is constructed. The user library is the scope of the users to be analyzed, and the user library includes multiple users, and all of them need to be numbered. And a user behavior database is established. The user behavior database is divided into two parts. The first part is the log record of the information browsed by the user, and the second part is the log record of the user's browsing information behavior. The two parts in the user behavior database share the same timeline.

[0022] A historical time period is preset, and the length of the preset historical time period is set according to actual needs. In order to accurately analyze users, the preset historical time period can be set longer. The preset historical time period includes multiple sampling moments, and the time interval between adjacent sampling moments is set according to actual needs.

[0023] In this embodiment, multiple users are analyzed. Taking any one user as an example, the following is the case.

[0024] Obtain the push information of the platform to the user within the preset historical time period, specifically: within the preset historical time period, the push information of the platform to the user at each sampling moment.

[0025] In an exemplary embodiment, each push message has a corresponding attribute, that is, type. For example: push information with the attribute of sports, push information with the attribute of cars, push information with the attribute of digital products, and so on. When the platform pushes information to the user at each sampling moment, it also obtains the information attributes of the push information to the user at each sampling moment within the preset historical time period, and obtains an information attribute vector according to the information attributes of the push information. Vectorizing the information attributes is to facilitate the calculation of the similarity between attributes in the following text. Then, step 1 further includes: obtaining a set of push information attribute vectors corresponding to the push information to the user within the preset historical time period. The set of push information attribute vectors includes the information attribute vectors of the push information to the user at each sampling moment within the preset historical time period, and the information attribute vectors in the set of push information attribute vectors are arranged in time sequence. Therefore, the push information to the user at each sampling moment corresponds to an information attribute vector.

[0026] The first part of the user behavior database includes user behavior data in time sequence, that is, within the preset historical time period, the push information of the platform to the user at each sampling moment and the information attribute vectors.

[0027] Then, according to the push information to the user, obtain the depth of the user's information cocoon. In an exemplary embodiment, as Figure 2 shown, the process of obtaining the depth of the information cocoon includes: Step 1-1: Obtain the similarity between the information attribute vectors at any two adjacent sampling moments, and obtain the similarity less than the preset similarity threshold to obtain the target similarity.

[0028] Since the information attribute vectors in the push information attribute vector set are arranged in time sequence, the similarity between any two adjacent sampling time information attribute vectors in the push information attribute vector set is obtained. Among them, the similarity between two vectors can be obtained by using existing similarity calculation methods, such as: cosine similarity, Pearson correlation coefficient, etc. It should be noted that after calculating the similarity, it needs to be normalized and brought into the numerical range of 0-1 for subsequent processing. In other exemplary embodiments, the DTW distance between two vectors can also be calculated, and then negative correlation normalization is performed to obtain the similarity.

[0029] It should be understood that the normalization method and negative correlation normalization method in this embodiment can be specifically set according to the actual situation. For example, normalization can adopt the maximum-minimum normalization method, or the following common methods: , represents the processing object, represents the exponential function with the natural constant as the base. Negative correlation normalization can adopt the following common methods: .

[0030] Taking the th user as an example, assuming that the preset historical time period includes sampling times, then, there are information attribute vectors, and similarities of information attribute vectors are obtained.

[0031] A preset similarity threshold is set, and the numerical range of the preset similarity threshold is 0-1. The specific value is set according to actual judgment needs, such as 0.8.

[0032] Compare the sizes of each similarity with the preset similarity threshold, obtain the similarities less than the preset similarity threshold, and define the similarities less than the preset similarity threshold as the target similarities.

[0033] Step 1-2: Use the positions between the two sampling times corresponding to each target similarity as the segmentation positions to segment the push information attribute vector set, and obtain multiple similarity segments.

[0034] Since the target similarity corresponds to two adjacent sampling moments, the positions between the two sampling moments corresponding to each target similarity are used as segmentation positions, and the set of push information attribute vectors is segmented according to each segmentation position to obtain multiple similar segments. Specifically: the first similar segment is the interval between the first sampling moment of the preset historical time period and the previous sampling moment among the two adjacent sampling moments corresponding to the first target similarity; the second similar segment is the interval between the latter sampling moment among the two adjacent sampling moments corresponding to the first target similarity and the previous sampling moment among the two adjacent sampling moments corresponding to the second target similarity; the third similar segment is the interval between the latter sampling moment among the two adjacent sampling moments corresponding to the second target similarity and the previous sampling moment among the two adjacent sampling moments corresponding to the third target similarity, and so on. The last similar segment is the interval between the latter sampling moment among the two adjacent sampling moments corresponding to the last target similarity and the last sampling moment of the preset historical time period.

[0035] Through the above segmentation method, the differences in the push information between different similar segments are relatively large, while the differences in the push information within the same similar segment are relatively small, and the push information is relatively similar.

[0036] Step 1-3: Based on each similar segment, obtain the information cocoon depth of the user.

[0037] In an exemplary embodiment, as Figure 3 shown, the process of obtaining the information cocoon depth of the user includes: Step 1-3-1: Obtain the mean value of the similarities included in each similar segment, and obtain the number of information attribute vectors in each similar segment.

[0038] It should be understood that each similar segment includes multiple sampling moments, that is, multiple similarities. Then, for any similar segment, obtain the mean value of the similarities included in the similar segment. At the same time, obtain the number of information attribute vectors in the similar segment, that is, the number of sampling moments included.

[0039] Step 1-3-2: According to the mean value of the similarities, the number of information attribute vectors, and the number of similar segments, obtain the information cocoon depth sub-features of each similar segment.

[0040] For any similar segment, according to the mean value of the similarities of the similar segment, the number of information attribute vectors of the similar segment, and the number of similar segments, obtain the information cocoon depth sub-feature of the similar segment. The information cocoon depth sub-feature is proportional to the mean value of the similarities and the number of information attribute vectors, and inversely proportional to the number of similar segments.

[0041] Step 1-3-3: Obtain the sum value of the information cocoon depth sub-features of each similar segment and normalize it to obtain the information cocoon depth of the user.

[0042] In an exemplary embodiment, the calculation formula for the information cocoon depth is given as follows: ; Wherein, represents the th user, represents the information cocoon depth of the th user, represents the th similar segment of the th user, represents the number of similar segments of the th user, represents the mean similarity of the th similar segment of the th user, represents the number of information attribute vectors of the th similar segment of the th user, represents the normalization function.

[0043] represents the information cocoon depth sub-feature of the th user.

[0044] For the th user, when the platform pushes information to the th user, the platform analyzes the similarity and the duration of similar information (i.e., the number of information attribute vectors of the similar segment) of the information pushed to the th user. The more similar the information pushed by the platform to the th user is, the longer the duration of the similar information is, and the fewer the total number of similar segments is, it indicates that the attribute (i.e., type) of the pushed information obtained by the th user is more unique. Then, the depth of being trapped in the information cocoon is higher. Therefore, use the product of the mean similarity and the length of the similar segment in the similar segments with higher similarity of the th user, and divide it by the total number of similar segments as the information cocoon depth sub-feature of each similar segment. Integrate the information cocoon depth sub-features of all similar segments to obtain the information cocoon depth of the th user.

[0045] Such as Figure 4As shown in the figure, it is a logical relationship diagram between the curve of the depth of the information cocoon and other related curves. The abscissa is time, and the ordinate is amplitude. Among them, curve B1 represents the curve of the change in the possibility of the user obtaining push information of other attributes, curve B2 is the similarity change curve, and curve B3 is the change curve of the depth of the user's information cocoon. Figure 4 Among them, as the similarity of the content pushed by the platform to the user is getting higher and higher, and the pushing time of similar attributes is getting longer and longer, the greater the depth of the information cocoon that the user receives when obtaining information on the platform, and the lower the possibility of the user receiving push information of other attributes.

[0046] By adopting the above method, the depth of the information cocoon of each user is obtained.

[0047] Step 2: According to the viewing duration of each push message of the user for a preset historical period, obtain the satisfaction degree of the user with the push message.

[0048] When the platform pushes information to the user, it will use the user's historical behavior habits to conduct user behavior analysis, and then push information that the user is more interested in. However, the user's acceptance threshold for push information of the same attribute will become higher and higher, resulting in too low accuracy of information push. Therefore, in this embodiment, by performing similarity analysis on the user's historical push information, it is judged whether the push information for the user is too similar and lasts for a long time, and combined with the user's historical viewing behavior of the information, it is used to obtain whether each user is satisfied with the information pushed by the platform, so as to describe the satisfaction degree of the user with the current push information of the platform.

[0049] In an exemplary embodiment, as Figure 5 shown, the process of obtaining the satisfaction degree includes: Step 2-1: Obtain the viewing duration of the user for each push message, and obtain the average viewing duration of all users for the same push message.

[0050] After the user obtains each push message of the system, the user will view each push message. Then, obtain the viewing duration of the user for each push message. The viewing duration is specifically the time length obtained by timing from when the user opens a certain push message to when the user closes the push message. For example: if the push message is a link to a sports news, then start timing from when the user opens the link to when the user closes the link, and the time length obtained is the viewing duration of the link. Thus, the viewing duration of the user for each push message within the preset historical period is obtained.

[0051] The second part of the user behavior database includes the viewing duration of the user for each push message within the preset historical period.

[0052] Since the system will push information to each user in the user library, there will be a situation where the same push information is pushed to multiple users, and multiple users will view the same push information. Then, obtain the viewing duration of multiple users for the same push information, and then calculate the average value to obtain the average viewing duration of all users for the same push information.

[0053] Step 2-2: Based on the difference between the viewing duration of the user for each push information and the average viewing duration of the push information, obtain the anti-fooling parameter of the user for the push information.

[0054] For any push information, according to the difference between the viewing duration of the user for the push information and the average viewing duration of the push information, obtain the anti-fooling parameter of the user for the push information. The purpose of setting the anti-fooling parameter is to prevent the user from mechanically viewing the push information in a non-human way. The specific behavior can be manifested as the user opens the push information, stays on the page of the push information for a long time, but does not actually receive the information.

[0055] In an exemplary embodiment, the calculation formula of the anti-fooling parameter is as follows: ; Wherein, represents the anti-fooling parameter of the th user for the th push information, represents the viewing duration of the th user for the th push information; represents the average viewing duration of all users for the th push information; represents the natural constant; represents the pi. The calculation formula of is essentially a variant of the Gaussian function, and the parameter is only

[0056] Step 2-3: According to the anti-fooling parameters of the user for each push information, the viewing duration of the user for each push information, and the average viewing duration, obtain the satisfaction degree of the user for the push information.

[0057] In an exemplary embodiment, the calculation formula of the satisfaction degree is as follows: ; Wherein, represents the satisfaction degree of the th user, represents the number of push information.

[0058] The normalization method for can be as follows: Obtain the for all push messages of the user, find the maximum and minimum values from them, and then use the maximum-minimum normalization method to normalize the for each push message of the user.

[0059] When the user views the push messages of the platform, they tend to invest more attention in the push messages they are interested in, while they usually only have less attention for the push messages they are not interested in. Intuitively, they invest more viewing time in the content they are interested in, and generally do not have a long viewing time for the content they are not interested in. Therefore, this embodiment uses this logic to calculate the satisfaction degree of the user for the push messages. The specific calculation logic is: The anti-fooling parameter is the weight of the difference between the viewing duration of the push message and its average viewing duration. The design logic of the anti-fooling parameter is: When the viewing duration of the push message is less than its average viewing duration by a larger margin, the anti-fooling parameter is larger, that is, the weight is larger. When the viewing duration of the push message exceeds the average viewing duration by a larger margin, the anti-fooling parameter is smaller. And, the smaller the average viewing duration, the faster the growth or decrease speed of the anti-fooling parameter, and vice versa.

[0060] As Figure 6 shown, it is a logical relationship diagram between the viewing speed of the user for the push message and the satisfaction degree. The abscissa is time, and the ordinate is amplitude. Among them, curve C1 is the viewing speed curve of the user for the push message, and curve C2 is the satisfaction degree curve of the user for the push message. The faster the viewing speed, the shorter the viewing duration and the lower the satisfaction degree.

[0061] By adopting the above process, the satisfaction degree of each user is obtained.

[0062] Step 3: Combine the information cocoon depth and the satisfaction degree to obtain the information recommendation degree of the user for the push message.

[0063] According to the obtained information cocoon depth and satisfaction degree of the user for the push message, the information recommendation degree of the user for the push message is obtained, which is used to describe whether the push message of the platform for the user meets the user's needs.

[0064] In an exemplary embodiment, the process of obtaining the information recommendation degree is as Figure 7 shown, including: Step 3-1: According to the information cocoon depth and the satisfaction degree, obtain the timely adaptability of the user for the push message.

[0065] The higher the information cocoon depth of the platform when pushing information to users, the lower the satisfaction degree of users with the pushed information, and the worse the timeliness adaptability between the users' expected pushed information viewing and the platform's pushed information. According to the information cocoon depth and satisfaction degree, the timeliness adaptability of users to the pushed information is obtained. The timeliness adaptability is inversely proportional to the information cocoon depth and directly proportional to the satisfaction degree.

[0066] In an exemplary embodiment, the calculation formula for the timeliness adaptability is as follows: ; Wherein, represents the timeliness adaptability of the -th user, represents the satisfaction degree of the -th user, represents the information cocoon depth of the -th user, represents the natural constant.

[0067] Step 3-2: Obtain the information recommendation degree according to the timeliness adaptability.

[0068] The lower the timeliness adaptability, the more necessary it is to recommend pushed information with new attributes to users. If the timeliness adaptability is higher, the less necessary it is to recommend information. Then, according to the timeliness adaptability, the information recommendation degree is obtained. The information recommendation degree is inversely proportional to the timeliness adaptability.

[0069] In an exemplary embodiment, the calculation formula for the information recommendation degree is: ; Wherein, represents the information recommendation degree of the -th user.

[0070] Step 4: Obtain the difference in the viewing durations of two adjacent times of the pushed information with the same attribute for the user, and obtain the information acceptance degree of the user for the pushed information of each attribute according to the overall situation of the difference.

[0071] For the -th user, among the pushed information pushed by the system to the -th user, some pushed information belongs to the same attribute. For example, there are multiple pushed information that are all sports-related information. Then, obtain the viewing durations of the -th user for each pushed information with the same attribute within a preset historical time period, and then calculate the difference in the viewing durations of two adjacent viewings of the pushed information with the same attribute for the -th user. Then, within the preset historical time period, the The overall situation of the viewing duration differences of the push messages for the same attribute by multiple users is obtained, and finally the information acceptance degrees of the push messages for each attribute by the users are obtained.

[0072] In an exemplary embodiment, the calculation formula for the information acceptance degree is as follows: ; ; where, represents the information acceptance degree of the push message for the -th user for the -th attribute, represents the expected parameter, represents the -th viewing of the push message for the -th attribute by the -th user, represents the total number of viewings of the push message for the -th attribute by the -th user, represents the viewing duration of the -th viewing of the push message for the -th attribute by the -th user, represents the viewing duration of the -th viewing of the push message for the -th attribute by the -th user.

[0073] The difference between the viewing duration of the last viewing of the push message for a certain attribute by the user and the viewing duration of the adjacent previous viewing represents the expected growth amount of the viewing expectation of the user for the push message of this attribute in this viewing compared with the previous viewing expectation of the user for the push message of this attribute. Then, all the expected growth amounts are added up. The larger the total expected growth amount, the higher the acceptance degree of the user for the push message of this attribute as the time sequence increases, and vice versa. The purpose of the expected parameter is that when the viewing expectation of the user for the push message of a certain attribute shows negative growth, the acceptance degree of the user for the push message of this attribute is forced to be 0.

[0074] Using the above method, the information acceptance degrees of the push messages for each attribute by the users can be obtained.

[0075] Step 5: Perform information recommendation for the users according to the information recommendation degree and the information acceptance degree.

[0076] In an exemplary embodiment, as Figure 8 shown, the specific process of performing information recommendation for the users includes: Step 5-1: Obtain the attributes of the push messages with an information acceptance degree greater than the preset information acceptance degree threshold, and obtain the target attributes.

[0077] Preset an information acceptance degree threshold, and this preset information acceptance degree threshold is within the numerical range of 0-1, and the specific value is set according to actual judgment needs.

[0078] Compare the information acceptance degree of the push messages of each attribute for the th user with the preset information acceptance degree threshold, obtain the push messages with an information acceptance degree greater than the preset information acceptance degree threshold, and define the attributes of the obtained push messages with an information acceptance degree greater than the preset information acceptance degree threshold as the target attributes.

[0079] Step 5-2: Multiply the information acceptance degree of each target attribute by the information recommendation degree to obtain the distribution weight of the push messages of each target attribute.

[0080] For any one of the target attributes corresponding to the th user, calculate the product of the information acceptance degree of this target attribute and the information recommendation degree of the th user, and use the product corresponding to this target attribute as the distribution weight of the push message of this target attribute for the th user.

[0081] Step 5-3: Push the push messages of each target attribute to the user according to the distribution weight of the push messages of each target attribute.

[0082] The system pushes the push messages of each target attribute to the th user according to the distribution weight of the push messages of each target attribute of the th user. Specifically: The system uses the distribution weight as the push ratio, and uses the push ratio of the distribution weight of the push messages of each target attribute of the th user to push the push messages of each target attribute to the th user.

[0083] In an exemplary embodiment, it is also possible to continue to push the push messages of the original attributes to the th user at the proportion of the timely adaptability of the th user. It should be understood that this part does not belong to a part of an information recommendation method based on artificial intelligence provided in this embodiment, and is not limited here.

[0084] Use the above process to push information to each user.

[0085] This embodiment also provides an information recommendation system based on artificial intelligence, including: a memory and a processor; the memory is connected to the processor, and the memory is used to store program instructions; the processor is used to implement the steps in the above-mentioned embodiment of the information recommendation method based on artificial intelligence when the program instructions are executed.

[0086] In an exemplary embodiment, the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned embodiment of the information recommendation method based on artificial intelligence are implemented.

[0087] It should be noted that: the above-mentioned sequence of embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0088] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. An information recommendation method based on artificial intelligence, characterized in that: include: Get the push information for the user within the preset historical time period, and obtain the user's information cocoon depth based on the push information for the user; Obtaining the user's satisfaction with the pushed information based on the user's reading time of each pushed information in a preset historical time period; Combining the information cocoon depth and satisfaction level, obtaining the user's information recommendation level for the pushed information; Obtain the difference between two consecutive reading times of the user for the push information of the same attribute, and obtain the user's acceptance degree of the push information of each attribute based on the overall situation of the difference; Information is recommended to the user according to the information recommendation degree and the information acceptance degree.

2. The information recommendation method based on artificial intelligence as claimed in claim 1, characterized in that: The information recommendation method based on artificial intelligence also includes: Obtaining a set of push information attribute vectors corresponding to the push information to the user within a preset historical time period, wherein the push information attribute vector set includes information attribute vectors of the push information to the user at each sampling moment within the preset historical time period; The method of obtaining the user's information cocoon depth based on the push information to the user includes: Obtaining the similarity of the information attribute vectors at any two adjacent sampling moments, and obtaining a similarity that is less than a preset similarity threshold, to obtain a target similarity; The position between two sampling moments corresponding to each target similarity is used as a segmentation position, and the push information attribute vector set is segmented to obtain a plurality of similar segments; Based on each similar segment, the depth of the user’s information cocoon is obtained.

3. The information recommendation method based on artificial intelligence as claimed in claim 2, characterized in that: Based on each similar segment, the user's information cocoon depth is obtained, including: Obtaining a mean value of the similarities contained in each similar segment, and obtaining the number of the information attribute vectors in each similar segment; According to the mean value of the similarity, the number of the information attribute vectors and the number of similar segments, the information cocoon depth sub-feature of each similar segment is obtained, wherein the information cocoon depth sub-feature is proportional to the mean value of the similarity and the number of the information attribute vectors, and inversely proportional to the number of similar segments; The sum of the information cocoon depth sub-features of each similar segment is obtained and normalized to obtain the user's information cocoon depth.

4. The information recommendation method based on artificial intelligence as claimed in claim 1, characterized in that: According to the reading time of each push information in the preset historical time period, the user's satisfaction with the push information is obtained, including: Obtain the user's reading time for each push message, and obtain the average reading time of the same push message for all users; Based on the difference between the reading time of each push information by the user and the average reading time of the push information, obtaining the foolproof parameter of the user for the push information; The user's satisfaction with the push information is obtained based on the user's foolproof parameters for each push information, the user's reading time for each push information, and the average reading time.

5. The information recommendation method based on artificial intelligence as claimed in claim 4, characterized in that: The calculation formula of the foolproof parameter is as follows: ； in, Indicates User to foolproof parameters for push messages, Indicates User to The reading time of each push message; Indicates that all users have Average reading time of push messages; represents a natural constant; represents pi; The calculation formula of the satisfaction degree is as follows: ； in, Indicates The satisfaction level of each user, Indicates the number of pushed messages. Represents the normalization function.

6. The information recommendation method based on artificial intelligence as claimed in claim 1, characterized in that: Combined with the information cocoon depth and satisfaction level, the user's information recommendation level for the pushed information is obtained, including: According to the information cocoon depth and satisfaction level, the user's timely adaptability to the pushed information is obtained, wherein the timely adaptability is inversely proportional to the information cocoon depth and directly proportional to the satisfaction level; According to the timely adaptability, the information recommendation degree is obtained, and the information recommendation degree is inversely proportional to the timely adaptability.

7. The information recommendation method based on artificial intelligence as claimed in claim 6, characterized in that: The calculation formula of the timely adaptability is as follows: ； in, Indicates Timely adaptability for each user, Indicates The satisfaction level of each user, Indicates The depth of a user's information cocoon, represents a natural constant; The calculation formula of the information recommendation degree is: ； in, Indicates The information recommendation level of a user.

8. The information recommendation method based on artificial intelligence as claimed in claim 1, characterized in that: The obtaining of the difference between two consecutive reading times of the push information of the same attribute by the user, and obtaining the information acceptance degree of the user for the push information of each attribute according to the overall situation of the difference, includes: The calculation formula of the information acceptance degree is as follows: ； ； in, Indicates User for The degree of acceptance of the information pushed by each attribute, represents the expected parameters, Indicates User to The first attribute of the push information Views: Indicates User to The total number of views of the push information of each attribute. Indicates User to The first attribute of the push information The reading time of each reading, Indicates User to The first attribute of the push information The reading time of each reading.

9. The information recommendation method based on artificial intelligence as claimed in claim 1, characterized in that: Recommending information to the user according to the information recommendation degree and the information acceptance degree, including: Acquire the attributes of the pushed information whose information acceptance degree is greater than a preset information acceptance degree threshold, and obtain the target attribute; Multiply the information acceptance degree of each target attribute by the information recommendation degree to obtain the distribution weight of the pushed information of each target attribute; The push information of each target attribute is pushed to the user according to the allocation weight of the push information of each target attribute.

10. An information recommendation system based on artificial intelligence, characterized in that: include: Memory and processor; The memory is connected to the processor; The memory is used to store program instructions; The processor is used to implement the artificial intelligence-based information recommendation method according to any one of claims 1 to 9 when the program instructions are executed.

Images