Diversity determination method and related device

Abstract

The invention discloses a diversity determination method and a related device, and the method comprises the steps: obtaining a content set comprising N contents and a content order of the content set, enabling the content order to represent a display order of the N contents when the content set is recommended, combining the N contents based on the ith content and the (i + 1) th content which are adjacent in the content order, and obtaining M first content combinations corresponding to the content set to determine what closely arranged contents in the N contents are. And finally, based on the content similarities corresponding to the M first content combinations, determining the content diversity of the content set. The content similarity of the first content combination can reflect the similarity degree between the two contents, so that whether similar contents are densely arranged or not can be measured, and whether the similar contents are densely arranged or not can influence the recommendation effect of the content set. On the basis, the content diversity can reflect whether similar contents are densely arranged or not, so that the content diversity is more accurately quantified.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a diversity determination method and related devices. Background Art

[0002] With the continuous development of Internet technology, there are more and more scenarios for recommending content to users. Specifically, content can be recommended to users through applications, web pages and other products. Correspondingly, users can interact with the recommended content through applications, web pages, etc. (such as viewing, liking, etc.). In this way, users can obtain content and the information conveyed by the content more conveniently.

[0003] In some recommendation scenarios, it is necessary to recommend a content set including multiple contents to users. In practical applications, for recommendation scenarios where content sets are recommended to users, the content diversity of the content set can be used as an indicator to measure the recommendation effect. Specifically, content diversity can reflect the distribution of multiple contents included in the content set. For example, if the content diversity is too low, it means that multiple contents are more likely to be of the same type. This may cause the content obtained by users to be limited to certain categories, resulting in an "information cocoon", which will affect the final recommendation effect.

[0004] In practical applications, the content diversity of a content set is usually quantified in order to measure the recommendation effect. However, the diversity determination method used in related technologies is difficult to accurately quantify the content diversity of a content set. Summary of the invention

[0005] In order to solve the above technical problems, the present application provides a diversity determination method and related devices, so that content diversity can reflect whether similar content is densely arranged, so that content diversity can be quantified more accurately.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] In one aspect, an embodiment of the present application provides a diversity determination method, the method comprising:

[0008] Acquire a content set including N contents, and acquire a content ranking of the content set; the content ranking is used to represent the display order of the N contents when recommending the content set, and N is an integer greater than 2;

[0009] Combining the i-th content and the i+1-th content that are adjacent to each other based on the content sorting of the N contents to obtain M first content combinations corresponding to the content set; i is an integer greater than or equal to 1 and less than N;

[0010] The content diversity of the content set is determined according to the content similarities respectively corresponding to the M first content combinations; the content similarity of the first content combination is determined based on the similarity between two contents in the first content combination.

[0011] In another aspect, an embodiment of the present application provides a diversity determination device, the device comprising an acquisition unit, a combination unit and a determination unit:

[0012] The acquisition unit is used to acquire a content set including N contents, and to acquire a content ranking of the content set; the content ranking is used to represent a display order of the N contents when recommending the content set, where N is an integer greater than 2;

[0013] The combining unit is used to combine the i-th content and the i+1-th content that are adjacent to each other in the content sorting of the N contents to obtain M first content combinations corresponding to the content set; i is an integer greater than or equal to 1 and less than N;

[0014] The determining unit is configured to determine the content diversity of the content set according to the content similarities respectively corresponding to the M first content combinations; the content similarity of the first content combination is determined based on the similarity between two contents in the first content combination.

[0015] On the other hand, an embodiment of the present application provides a computer device, the computer device comprising a processor and a memory:

[0016] The memory is used to store a computer program and transmit the computer program to the processor;

[0017] The processor is configured to execute the method described in any one of the preceding aspects according to instructions in the computer program.

[0018] On the other hand, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program. When the computer program is executed by a computer device, the computer device executes the method described in any of the above aspects.

[0019] On the other hand, an embodiment of the present application provides a computer program product, including a computer program, which, when executed on a computer device, enables the computer device to execute the method described in any one of the aforementioned aspects.

[0020] It can be seen from the above technical solution that first, a content set including N contents can be obtained, and the content ranking of the content set can be obtained, where N is an integer greater than 2. Among them, the content ranking can be used to characterize the display order of N contents when recommending the content set, so the N contents can be combined based on the adjacent i-th content and i+1-th content based on the content ranking to obtain M first content combinations corresponding to the content set, where i is an integer greater than or equal to 1 and less than N. Based on this, it can be determined what the closely arranged contents are among the N contents. Finally, the content diversity of the content set can be determined based on the content similarities corresponding to the M first content combinations. Among them, the content similarity of the first content combination can be determined based on the similarity between two contents in the first content combination, which can reflect the similarity between the two contents, and since the display order of the two contents is adjacent, the content similarity based on the first content combination can measure whether similar contents are densely arranged, and whether similar contents are densely arranged will affect the recommendation effect of the content set. Based on this, the content diversity can reflect whether similar contents are densely arranged, so that the content diversity can be quantified more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technical members in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0022] Figure 1 A schematic diagram of a user interface of an information flow product provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of an application scenario of a diversity determination method provided in an embodiment of the present application;

[0024] Figure 3 A flowchart of a diversity determination method provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of determining the category of an article based on a neural network classifier provided in an embodiment of the present application;

[0026] Figure 5 A schematic diagram of determining the category of text based on content similarity provided in an embodiment of the present application;

[0027] Figure 6 A schematic diagram of a Lorentz curve provided in an embodiment of the present application;

[0028] Figure 7 A distribution diagram of a content set provided in an embodiment of the present application;

[0029] Figure 8 A schematic diagram of an internal analysis iteration process of a target product provided in an embodiment of the present application;

[0030] Fig. 9 A recommendation schematic diagram based on content diversity provided in an embodiment of the present application;

[0031] Fig.10 A structural diagram of a diversity determination device provided in an embodiment of the present application;

[0032] Fig.11 A structural diagram of a terminal provided in an embodiment of the present application;

[0033] Fig.12 A structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0034] The embodiments of the present application are described below in conjunction with the accompanying drawings.

[0035] With the continuous development of Internet technology, content can be recommended to users through applications, web pages and other products, so that users can more conveniently obtain content and the information conveyed by the content. In some recommendation scenarios, it is necessary to recommend a content set including multiple contents to users. In actual applications, the content diversity of a content set can reflect the distribution of multiple contents included in the content set. The distribution of multiple contents is different, and the content diversity will be different. After being recommended to users, the experience brought to users will also be different, affecting the final recommendation effect. For example, if the content diversity is too low, it means that multiple contents are more likely to be of the same type. This will cause the content obtained by users to be limited to certain categories, resulting in an "information cocoon". If the content diversity is too high, it means that multiple contents are relatively scattered, which may make it difficult for users to obtain the content they want.

[0036] For ease of understanding, let's take the recommendation scenario of using information flow products to recommend content collections to users as an example. Information flow products are products that display content to users in the form of information flow. Usually, information flow products are presented in a form similar to social media on the user interface. For example, Figure 1 As shown, Figure 1The user interface diagram of an information flow product is shown. The information flow product can be run on a computer device 100 (such as a smart phone). The user interface of the information flow product can be provided with three sections: "Follow", "Discover" and "My". Among them, the "Follow" section is mainly used to display the content that the user follows, the "My" section is mainly used to display the content related to the user account, and the "Discover" section can be used to display content collections for users to facilitate users to discover new and interesting content. For example, in Figure 1 In the user interface shown, four contents can be displayed to the user at a time, so in this example, the content set can include four contents. In the information flow product, the user can scroll to obtain more content, such as the user can Figure 1 Scroll through the user interface shown to obtain more content. It is understood that the content collection may also include the four contents currently shown in the user interface and more content that can be obtained by scrolling through. It is understood that if Figure 1 If the four displayed contents are similar to the contents that the user is interested in, then the user will see contents similar to the contents that he is interested in. Therefore, it can be considered that the user is trapped in an information filter bubble. The "information cocoon" formed leads to information limitations and narrows the user's interests. Figure 1 The four contents shown are published by different authors and belong to different fields or different themes. Users can obtain more extensive information by accessing these different contents, which is helpful to help users expand their areas of interest and discover new and interesting contents.

[0037] In practical applications, quantifying content diversity can guide the recommendation process, such as preventing the formation of "information cocoons" that trap users in information filtering bubbles, and helping users expand their areas of interest. To this end, the content diversity of content collections is usually quantified.

[0038] In the related art, the screen dominance index or the cocoon index is used to evaluate content diversity, where the screen dominance index can refer to more than n similar contents among N contents. Obviously, the larger the value of the screen dominance index, the more concentrated the similar contents are, and the worse the experience after recommendation. Setting different parameters N and n will eventually determine completely different content diversity. The setting of parameters is usually artificially set based on experience, and the subjective deviation is large, resulting in poor accuracy of content diversity. The cocoon index is similar to the screen dominance index. It refers to the highest proportion of similar content among N contents, so it is also called the similarity proportion index. For example, a content set of N=10 is recommended to the user, of which 5 contents are similar, then the value of the cocoon index is 0.5. Compared with the screen dominance index, the cocoon index omits parameter settings. However, the cocoon index can only reflect the situation with the highest similarity proportion, and it is difficult to reflect the situation of other contents, which makes it difficult for the determined content diversity to accurately reflect the overall diversity of N contents. For example, for two content sets with N=10, the value of the cocoon index is 0.5. However, the other five contents in the first content set are not similar, while three of the other five contents in the second content set are similar. Obviously, the content diversity of the first content set is higher than that of the second content set, but the content diversity determined based on the cocoon index is the same.

[0039] It can be seen that the diversity determination method used in the related art is difficult to accurately quantify the content diversity of the content set. To this end, the embodiment of the present application provides a diversity determination method. For a content set including N contents, since the display order of the N contents is different when recommending the content set, the difference in the display order brings the user an intuitive experience of whether the user will see similar content in succession, which will directly affect the final recommendation effect. Therefore, in the present application, the content sorting of the content set is combined, that is, the display order of the N contents is considered, and the i-th content and the i+1-th content adjacent to each other based on the content sorting are combined in the N contents to obtain the M first content combinations corresponding to the content set, so that it can be determined what the closely arranged content in the N contents is. Then, the content diversity of the content set can be determined based on the content similarity corresponding to the M first content combinations. Among them, the content similarity of the first content combination can be determined based on the similarity between the two contents in the first content combination, which can reflect the similarity between the two contents, and since the display order of the two contents is adjacent, the content similarity based on the first content combination can measure whether there is a situation where similar content is densely arranged. Compared with the related art, the content diversity determined by the present application can reflect whether similar contents are densely arranged among N contents, thereby quantifying the content diversity more accurately.

[0040] The diversity determination method provided in the embodiment of the present application can be implemented by a computer device, which can be a terminal or a server, wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Terminals include but are not limited to smart phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, etc. The terminal and the server can be directly or indirectly connected via wired or wireless communication, and this application is not limited here. The embodiment of the present application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, audio and video, assisted driving, etc. The embodiment of the present application can be specifically applied to various recommendation scenarios, for example, video recommendation scenarios based on information flow products, product recommendation scenarios, etc.

[0041] It should be noted that in the specific implementation of the present application, the process of determining diversity may involve relevant data such as user information. When the above embodiments of the present application are applied to specific products or technologies, the user's separate consent or separate permission is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0042] Figure 2 The application scenario of the diversity determination method provided in the embodiment of the present application is shown. Figure 2 In the scenario shown, the server 200 is used as an example of the aforementioned computer device for illustration:

[0043] First, the server 200 can obtain a content set including N contents, and can obtain the content ranking of the content set, where N is an integer greater than 2. Among them, the content ranking can be used to characterize the display order of N contents when recommending a content set. When actually recommending a content set, the difference in display order brings the user an intuitive experience of whether the user will see similar content in succession, which will directly affect the final recommendation effect. Based on the content ranking, the content set composed of N contents can be considered as an ordered sequence, which can be specifically arranged as the 1st content, ..., the i-th content, the i+1th content, ..., the N-th content, where i is an integer greater than or equal to 1 and less than N.

[0044] Next, the server 200 can combine the i-th content and the i+1-th content that are adjacent to each other in the content sorting of the N contents to obtain M first content combinations corresponding to the content set. It can be seen that the first content combination can be composed of two adjacent contents. Based on the first content combination, it can be determined what the closely arranged contents in the N contents are, which can reflect the contents that the user will see in succession. Among them, M is used to represent the number of first content combinations composed of two adjacent contents in the N contents, so the value of M is a positive integer less than N.

[0045] Finally, the server 200 can determine the content diversity of the content set based on the content similarities corresponding to the M first content combinations. Among them, the content similarity of the first content combination can be determined based on the similarity between the two contents in the first content combination, and can reflect the degree of similarity between the two contents. Since the display order of the two contents in the first content combination is adjacent, the content similarity based on the first content combination can measure whether similar contents are densely arranged, and whether similar contents are densely arranged affects whether users will see similar contents in succession, thereby affecting the recommendation effect of the content set. Compared with the related art, the use of this application can make the content diversity reflect whether similar contents are densely arranged, so that the content diversity can be quantified more accurately.

[0046] Figure 3 A flowchart of a diversity determination method provided in an embodiment of the present application is provided, and a server is used as an example of the aforementioned computer device for illustration. The method includes S301-S303:

[0047] S301: Obtain a content set including N contents, and obtain a content ranking of the content set.

[0048] When it is necessary to evaluate content diversity, the server can obtain a content set including N contents, and can obtain the content ranking of the content set. Among them, the content set may refer to a set whose content diversity needs to be evaluated, and the content ranking can be used to characterize the display order of N contents when recommending a content set. In practical applications, for the case where N is equal to 2, since there are only two contents, the order of display does not have a very intuitive impact on the user experience. Therefore, when the present application determines the content diversity in combination with the content ranking, it is more targeted at the case where N is greater than 2, so N is an integer greater than 2. Based on this, by distinguishing the scenarios where N is greater than 2 or N=2, it is beneficial to improve the saving of computing resources. In addition, the present application does not impose any restrictions on the form of the N contents. For example, it can be in the form of videos, articles, pictures, etc.

[0049] S302: Combine the i-th content and the i+1-th content that are adjacent to each other based on the content sorting of N contents to obtain M first content combinations corresponding to the content set.

[0050] Among them, content sorting can be used to characterize the display order of N contents when recommending a content set. When actually recommending a content set, the difference in display order brings the user an intuitive experience of whether the user will see similar content in succession, which will directly affect the final recommendation effect. Therefore, in the embodiment of the present application, the content diversity of the content set will be determined in combination with the content sorting. Specifically, the N contents can be combined based on the i-th content and the i+1-th content that are adjacent to each other in the content sorting to obtain M first content combinations corresponding to the content set, where i is an integer greater than or equal to 1 and less than N. It can be seen that the first content combination can be composed of two adjacent contents. Based on the first content combination, it can be determined what the closely arranged contents among the N contents are, which can reflect the content that the user will see in succession.

[0051] It can be understood that M is used to represent the number of first content combinations formed by combining two adjacent contents in N contents, so the value of M is a positive integer less than N. It should be noted that this application does not make any restrictions on the specific value of M, and it can be flexibly set according to the actual needs for the accuracy of content diversity. For example, setting M first content combinations as all first content combinations of the content set can comprehensively consider the situation of all adjacent combinations, which is conducive to determining more accurate content diversity. M first content combinations can also be set as the first content combinations of part of the content set, which is conducive to saving computing resources, but at the same time considering some first content combinations, so compared with the related art, the accuracy of content diversity can still be improved. For example, when N=4, there are 3 first content combinations composed of two adjacent contents, and M=1, M=2 or M=3 can be set accordingly. If M=3 is set, all first content combinations can be comprehensively considered, which is conducive to determining more accurate content diversity. If M=1 or M=2 is set, it is conducive to saving computing resources, but at the same time considering some first content combinations, so compared with the related art, the accuracy of content diversity can still be improved.

[0052] S303: Determine the content diversity of the content set according to the content similarities corresponding to the M first content combinations.

[0053] When actually recommending a content set, closely arranged content can reflect the content that the user will see in succession. Therefore, after determining M first content combinations, the server can determine the content diversity of the content set based on the content similarities corresponding to the M first content combinations. Among them, the content similarity of the first content combination is determined based on the similarity between two contents in the first content combination, and can reflect the degree of similarity between the two contents. Since the display order of the two contents in the first content combination is adjacent, the content similarity based on the first content combination can measure whether similar content appears densely arranged, and whether similar content is densely arranged affects whether the user will see similar content in succession, thereby affecting the recommendation effect of the content set. Compared with the related art, the use of this application can make the content diversity reflect whether similar content appears densely arranged, so that the content diversity can be quantified more accurately.

[0054] Based on this, the present application measures whether similar content appears densely arranged by combining content sorting and content similarity, so that the determined content diversity has the ability to reflect the content exposure density, which can be more accurate. For ease of understanding, taking N=4 as an example, these 4 contents can include two A contents, one B content and one C content, which can constitute two different content sets such as AABC and ABCA under different content sorting. The difference between the two is that there is a difference in the order in which content A appears (which can also be considered as the time of appearance), which makes the exposure density of content A different. For AABC, the two A contents are densely arranged, and it is considered that dense exposure has occurred, while for ABCA, the two A contents are not densely arranged, and it can be considered that there is no dense exposure. After recommendation, the experience brought to users is different. Accordingly, AABC can determine a content diversity value lower than ABCA. It can be seen that the present application measures whether similar content appears densely arranged by comprehensive content sorting, so that the determined content diversity has the ability to reflect the content exposure density, which can be more accurate.

[0055] The present application does not limit the method for determining the content similarity of the first content combination. For ease of understanding, the present application provides the following method as an example:

[0056] Content similarity can reflect the similarity between two contents. In practical applications, the similarity between two contents can be determined based on the content similarity between the two contents. Therefore, in one possible implementation, the content similarity can be determined based on the content similarity between the two contents. That is, determining the content similarity from the difference between the contents themselves can accurately reflect the similarity between the two contents. Usually, the contents have corresponding categories, which can indicate the subject, type, field, etc. of the contents. The difference between the categories of two contents can also reflect the similarity between the two contents. Therefore, in another possible implementation, the content similarity can be determined based on the category similarity between the two contents.

[0057] It should be noted that the present application does not impose any limitation on how to determine the content diversity according to the M first content combinations. For ease of understanding, the present application embodiment provides the following method as an example:

[0058] The first content combination is composed of two adjacent contents, which can most directly reflect the content that the user will see in succession. When actually recommending a content set, in addition to the first content combination, two contents that are not adjacent but arranged at intervals can also reflect the situation between the contents seen by the user. For example, two similar contents are arranged at an interval, which gives the user a feeling similar to that when similar contents are seen in succession. Therefore, in a possible implementation method, on the basis of M first content combinations, a second content combination composed of two non-adjacent contents in N contents can also be integrated, so as to more comprehensively measure content diversity and help improve the accuracy of content diversity. In specific implementation, the server can combine the i-th content and the j-th content of N contents based on content sorting to obtain K second content combinations corresponding to the content set, where j is an integer greater than i+1 and less than or equal to N. It can be seen that the two contents in the second content combination are not arranged adjacently, but arranged at intervals. Accordingly, when the aforementioned S303 is specifically implemented, the server can determine the content diversity of the content set according to the content similarities corresponding to the R target content combinations, wherein the R target content combinations can include M first content combinations and K second content combinations. Similar to the first content combination, the content similarity of the second content combination can be determined based on the similarity between the two contents in the second content combination, which can reflect the similarity between the two contents in the second content combination. Based on this, the K second content combinations can be integrated on the basis of the M first content combinations, so that the content diversity can be measured more comprehensively, which is conducive to improving the accuracy of the content diversity.

[0059] It is understandable that K is used to represent the number of second content combinations formed by combining two separated contents in N contents, so the value of K is a positive integer less than N. It should be noted that this application does not impose any restrictions on the specific value of K, and it can be flexibly set according to the actual needs of the accuracy of content diversity, such as setting K with reference to the setting method of M. It is understandable that for the value of R, firstly, R = M + K, and secondly, since any two contents in N contents are combined, the maximum value of R is It can represent the total number of combinations of two contents from N contents.

[0060] It should be noted that the present application does not impose any limitation on how to determine the content diversity according to the R target content combinations. For ease of understanding, the present application embodiment provides the following method as an example:

[0061] When recommending a content set in practice, the display order of similar content is different, and the user's experience is also different. For example, when the display order of two similar contents is separated by one digit, the user's experience of seeing similar content in a row will be stronger, and the experience is poor. If the display order of two similar contents is separated by ten digits, the user's experience of seeing similar content in a row will be weaker, and the experience is better. Therefore, in a possible implementation, the content diversity can be determined in combination with the distance between the two contents. In specific implementation, first, the server can determine the distance factor of the target content combination based on the ranking difference of the two contents in the content ranking in the target content combination. The distance factor can characterize the distance between the two contents, and specifically can reflect the distance between the display order of the two contents. Based on this, for each target content combination, there is a corresponding distance factor to reflect the distance factor of the distance between the two contents included. It can be understood that the same two contents have different distance factors, and the degree of influence on content diversity is different. Therefore, when determining content diversity, the server can determine the content diversity of the content set according to the content similarity and distance factor corresponding to the R target content combinations. Based on this, we can comprehensively consider the distance between the content display order on the basis of content similarity, so as to more accurately measure content diversity, which is conducive to improving the accuracy of content diversity.

[0062] Among them, the present application does not make any limitation on how to determine the distance factor of the target content combination. It can be understood that the distance factor reflects the distance between the display order of two contents, so in a possible implementation, the ranking difference |ij| can be directly determined as the distance factor, so that the distance factor is positively correlated with the ranking difference, and the size of the distance factor can intuitively reflect the distance of the interval. In addition, in another possible implementation, the reciprocal of the ranking difference 1 / |ij| can also be determined as the distance factor, so that the distance factor is negatively correlated with the ranking difference, and the size of the distance factor can reflect the degree of influence of the distance of the interval on the content diversity, such as, the farther the interval, the smaller the distance factor, indicating that the impact of this target content combination on the content diversity is relatively small. It can be understood that when the target content combination is the first content combination, that is, the case of j=i+1, since the two contents in the first content combination are adjacent, the distance factors of the M first content combinations are the same. Taking the distance factor as |ij| or the distance factor as 1 / |ij| as an example, the distance factors of the M first content combinations are all 1.

[0063] Furthermore, the present application does not make any limitation on how to determine the content diversity according to the content similarity and distance factor corresponding to the R target content combinations. For ease of understanding, the present application embodiment provides the following method as an example:

[0064] For the target content combination, the content similarity can reflect the similarity between the two contents in the target content combination, and the distance factor can reflect the distance between the two contents. Therefore, the two dimensions of comprehensive content similarity and distance factor can be used to measure the degree of influence of the target content combination on the content diversity of the content set as a whole. Therefore, in a possible implementation method, the degree of influence of the target content combination on the content diversity can be determined first, and then the content diversity of the content set can be determined based on the degree of influence of each of the R target content combinations on the content combination. In specific implementation, the server can determine the diversity factor of the target content combination for the content diversity based on the content similarity of the target content combination and the distance factor of the target content combination. Based on this, the diversity factor is determined by combining the two dimensions of content diversity and distance factor. The diversity factor can be used to characterize the degree of influence of the target content combination on the content set as a whole. After determining the diversity factor of each target content combination, the server can determine the content diversity of the content set based on the diversity factors corresponding to the R target content combinations. Based on this, a method for determining diversity with target content combinations as processing units is provided. Determining the diversity factor of the target content combination is conducive to more accurately measuring the impact of each target content combination on content diversity, and then comprehensively determining the content diversity based on the impact of R target content combinations, which is conducive to improving the accuracy of content diversity.

[0065] For a better understanding, take the target content combination composed of the ith content and the jth content as an example, and take the distance factor as the aforementioned 1 / |ij| as an example. If the content similarity of the target content combination is high, it indicates that the ith content and the jth content are similar content. At the same time, if the distance factor of the target content combination is small, it indicates that the ith content and the jth content are far apart. It can be considered that although the two contents are similar, the impact on content diversity is relatively small due to the long distance. For the case where the distance factor is large, the impact on content diversity is relatively large due to the close distance. In this way, the diversity factors determined for different situations are different, but they are all determined by comprehensively considering the content similarity and the distance factor, and can accurately reflect the degree of influence of the target content combination on content diversity. For further understanding, the diversity factor of the target content combination can be determined by the following formula:

[0066]

[0067] In the above formula, r is an integer greater than or equal to 1 and less than or equal to R, which is used to represent the rth target content combination, D r It can represent the diversity factor of the rth target content combination, C i Can represent the i-th content, C j can represent the jth content, similarity(C i , C j ) can represent content similarity, where i is an integer greater than or equal to 1 and less than N, and j is an integer greater than or equal to i+1 and less than or equal to N.

[0068] Through the above embodiments, the diversity factors of target content combinations are illustrated. In order to better understand the method of determining content diversity based on the diversity factors of R target content combinations provided by the present application, the above example D r Let’s take the diversity factor of the rth target content combination as an example:

[0069] In practical applications, after determining the diversity factor of each target content combination, the diversity factors of the R target content combinations can be summed to represent the degree of influence of the R target content combinations on the content diversity of the content set. That is, taking all content combinations corresponding to N content as an example, the content diversity can be quantified more comprehensively and accurately. Specifically, it can be determined by the following formula:

[0070]

[0071] In the above formula, Y0 can represent the sum of the diversity factors of R target content combinations.

[0072] Understandably, D r The determination method can also be considered as using the distance factor to correct the content similarity, and determining the corrected content similarity as the diversity factor. In practical applications, the value of content similarity is usually [0,1]. Therefore, in order to ensure that the value of the sum of the diversity factors of the R target contents is still [0,1], the sum of the distance factors of the R target content combinations can be used for normalization. Correspondingly, it can be determined by the following formula:

[0073]

[0074] In the above formula, Y can represent the result after normalizing Y0.

[0075] In the above formula for diversity factor, similarity(C i , C j ) can represent content similarity. It is understandable that the content similarity can be determined in different ways. i , C j ) can be expressed in different forms. For example, in one possible implementation, taking the aforementioned content similarity as being determined based on the category similarity between two contents, when using the difference between categories to reflect the similarity between the contents, for the sake of simplicity, the contents of the same category can be considered to be the same (the similarity can be considered to be 1), and the contents of different categories can be considered to be different (the similarity can be considered to be 0), which can help simplify the calculation and save computing resources. Accordingly, the content similarity can be determined by the following formula:

[0076]

[0077] In the above formula, x i Can represent the category of the i-th content, x j It can represent the category of the jth content, Can be an indicative function, used to represent x i and x j Whether they are equal or not, the corresponding values ​​are as follows:

[0078]

[0079] For another example, in another possible implementation, taking the aforementioned content similarity as being determined based on the content similarity between two contents, it is usually possible to extract features from the content to obtain a feature vector, and use the feature vector to represent the content itself. The content similarity between the two contents can also be represented by the cosine similarity between the feature vectors of the two contents. Accordingly, the content similarity can be determined by the following formula:

[0080] similarity(C i , C j )=cosin(v i , v j )

[0081] In the above formula, v i The feature vector that can represent the i-th content, v j The feature vector that can represent the j-th content.

[0082] For better understanding, the present application embodiment takes the aforementioned article as an example and provides a method for determining the category of the article as an example. For details, please refer to Figure 4 As shown, for any newly stored article, the article can be input into the Large Language Model (LLM), and the feature vector of the article can be obtained by using the LLM to extract the features of the article. Then, the feature vector of the article can be input into the neural network classifier, and the category of the article can be determined by the neural network classifier. In practical applications, the category can be a multi-valued label. Assuming that the multi-valued labels of two contents have an intersection, the two articles are considered similar. In practical applications, the feature vector of the article can also be stored in the feature vector library, so that when the content diversity is determined later, the feature vector of the article can be directly obtained to determine the content similarity between the two articles.

[0083] It can be seen from the above technical solution that first, a content set including N contents can be obtained, and the content ranking of the content set can be obtained, where N is an integer greater than 2. Among them, the content ranking can be used to characterize the display order of N contents when recommending the content set, so the N contents can be combined based on the adjacent i-th content and i+1-th content based on the content ranking to obtain M first content combinations corresponding to the content set, where i is an integer greater than or equal to 1 and less than N. Based on this, it can be determined what the closely arranged contents are among the N contents. Finally, the content diversity of the content set can be determined based on the content similarities corresponding to the M first content combinations. Among them, the content similarity of the first content combination can be determined based on the similarity between two contents in the first content combination, which can reflect the similarity between the two contents, and since the display order of the two contents is adjacent, the content similarity based on the first content combination can measure whether similar contents are densely arranged, and whether similar contents are densely arranged will affect the recommendation effect of the content set. Based on this, the content diversity can reflect whether similar contents are densely arranged, so that the content diversity can be quantified more accurately.

[0084] Through the above embodiments, the diversity determination method provided by the present application is described in detail. For a better understanding, the present application embodiment provides the following method as another example for determining content diversity according to the content similarities corresponding to the M first content combinations:

[0085] In practical applications, content diversity can also be decomposed into two dimensions: richness and balance. Among them, richness can reflect the number of types of N content that users see. In practical applications, the higher the number of types of N content, the higher the richness of the content set. For example, if N=10, if the number of types of the 10 content categories is 2 (such as sports and society), then the richness of the content set can be considered to be 2. If the number of types of the 10 content categories is 5 (such as sports, society, fitness, travel and food), then the richness of the content set is 5. In the recommendation scenario, richness is positively correlated with user experience. For example, after recommendation, a content set with a richness of 5 can bring users more new information, which is conducive to expanding new interests, etc., while a content set with a lower richness is difficult to expand user interests after recommendation, and is prone to information limitation. For the recommendation system, recommending a content set with higher richness is conducive to continuously expanding users' new interests, etc., thereby improving user activity and improving recommendation effects. Balance can reflect the difference between the number of contents in each category that users see. In actual applications, the number of contents in some categories among N contents is significantly higher than that in other categories, indicating that the difference between the number of contents in different categories is more significant, and the balance is considered to be low. In other words, the greater the difference between the number of contents in each category, the lower the balance. In the recommendation scenario, balance is positively correlated with user experience. The lower the balance, the more users frequently see the same type of content, which easily creates a sense of density and repetition, and is also prone to information limitation.

[0086] Therefore, in a possible implementation, the content diversity can be determined by comprehensively considering the richness, balance, and content similarity of the M first content combinations, so that the determined content diversity can reflect the comprehensive situation of the richness, balance, and whether similar content is densely arranged of the N content, and has higher accuracy. In specific implementation, for the content set, the number of categories m (m is an integer greater than or equal to 1 and less than or equal to N) corresponding to the N content can be counted based on the categories of the N content, and the number of content under each category can be counted. In this way, the number of categories can reflect the richness, and the number of content under each category can reflect the balance. Therefore, after determining the number of categories and the number of content under each category, the content diversity can be determined in combination with the content similarity of the M first content combinations. Based on this, the determined content diversity can reflect the comprehensive situation of the richness, balance, and whether similar content is densely arranged of the N content, and has higher accuracy.

[0087] It should be noted that the present application does not limit the method for determining content diversity based on the number of contents, the number of categories m, the number of contents under each category, and the content similarity of the M first content combinations. For ease of understanding, the present application provides the following method as an example:

[0088] It is understandable that, for N content categories, in a possible implementation, the aforementioned Figure 4 The method shown in the figure can be used to determine, that is, the categories of N content can be already determined. For this, the number of categories and the number of content under each category can be directly counted, which is conducive to improving efficiency. In practical applications, if the similarity between two contents is high, the two contents can be considered to be of the same type. Therefore, in another possible implementation method, the categories of the two contents can also be determined based on the content similarity between the two contents. That is, the categories of N content can be determined based on content similarity when needed. For example, taking the aforementioned content as an article as an example, you can refer to Figure 5 As shown, first, the exposed articles can be collected (see S501), that is, N articles can be determined, and then the feature vectors of the articles can be queried from the feature vector library (see S502), and then the content similarity between the articles can be calculated based on the feature vectors (see S503), for example, the cosine similarity between the feature vectors can be calculated as the content similarity, and finally the categories corresponding to the N articles can be determined based on the content similarity (see S504), based on which, the category of the content can be determined based on the content similarity.

[0089] For ease of understanding, the embodiment of the present application takes determining the category of content based on content similarity as an example, and provides another example description of the method of determining content diversity in the present application:

[0090] In practical applications, if the aforementioned M first content combinations are all first content combinations of the content set, and the aforementioned content similarity is determined based on the content similarity between two contents, based on this, the categories of N contents can be determined based on the content similarity of the M first content combinations. In specific implementation, the server can determine the N categories corresponding to the N contents according to the content similarities corresponding to the M first content combinations, and count the number of categories corresponding to the N categories, wherein the N categories can be used to distinguish the N contents. In practical applications, the N categories can be arranged according to the category sorting, and the category sorting is consistent with the content sorting, that is, the category sorting can reflect the display order when the content is represented by categories.

[0091] Since the N categories are determined based on content similarity, and content similarity is determined based on the content similarity between the contents themselves, in this implementation, two contents with the same category can be considered to be the same content, which is conducive to simplifying the calculation process. Therefore, when the number of categories is greater than or equal to 2 and less than N, it can be considered that there are the same contents in these N contents, but they are not all the same. Therefore, at this time, the content diversity can be measured based on the category, and no longer pay attention to what the content itself is, which is conducive to improving efficiency. Specifically, if the number of categories is greater than or equal to 2 and less than N, the content diversity of the content set can be determined based on the number of contents, the number of categories, the number of contents under each category, and the category sorting of the content set. Based on this, on the basis of coverage richness and balance, for the case where there is the same content but not all the same, the same content is measured by combining the category sorting method to measure whether it is densely arranged, thereby improving the accuracy of content diversity.

[0092] Correspondingly, if the number of categories is equal to 1 or the number of categories is equal to N, it can be considered that the N contents are the same or the N contents are different. At this time, it can be simply considered that the content diversity is the worst when the number of categories is 1, and the content diversity is the highest when the number of categories is N. Considering that two contents of the same category may also have differences, in this case, even if the number of categories is equal to 1 or the number of categories is equal to N, different sorting conditions will result in a dense arrangement of more similar contents. Therefore, for the case where the number of categories is equal to 1 or the number of categories is equal to N, attention can be paid to the content similarity between two adjacent contents. Specifically, the content diversity of the content set can be determined based on the number of contents, the number of categories, the number of contents under each category, and the content similarities corresponding to the M first content combinations. Based on this, for the case where the number of categories is equal to 1 or the number of categories is equal to N, the present application can also be used to determine a more accurate content diversity, thereby improving the applicability of the present application.

[0093] Based on this, the method of using content similarity to determine the categories of N contents can be divided into two different situations by counting the number of types: the number of types is greater than or equal to 2 and less than N, and the number of types is equal to 1 or the number of types is equal to N. Different content diversity determination methods are provided for different situations, but they can all achieve comprehensive consideration of whether similar content is densely arranged, thereby ensuring that different determination methods can have high accuracy.

[0094] In the case where the number of categories is greater than or equal to 2 and less than N, this application does not limit how to determine content diversity based on the number of contents, the number of categories, the number of contents under each category, and the category ranking. For better understanding, the embodiments of this application provide the following methods as examples:

[0095] The number of categories can reflect the richness, and the number of contents under each category can reflect the balance. Therefore, in a possible implementation method, the initial diversity of the content set can be determined based on the number of contents, the number of categories, and the number of contents under each category of the content set. The initial diversity obtained based on this is obtained by combining the two dimensions of richness and balance, and can preliminarily measure the content diversity. Then, the initial diversity can be corrected by category sorting, so as to comprehensively consider whether there are dense arrangements of the same categories, and further improve the accuracy of content diversity. Specifically, if the categories of the same category in N categories are adjacently arranged based on category sorting, it indicates that the same categories are densely arranged, that is, the same content is densely arranged, which will lead to the user seeing the same content in a row. The bad experience, so when correcting, can be adjusted downward to obtain the content diversity of the content set. Based on this, the two-step analysis method can be used to first determine the initial diversity obtained by combining the two dimensions of richness and balance, and then combine the category sorting to consider whether the same content is densely arranged, so as to improve the accuracy of content diversity.

[0096] Through the above embodiments, the method of determining content diversity by combining richness, balance and content similarity of the first content combination in this application is described in detail. It should be noted that in this application, the overall result (such as the aforementioned initial diversity) is obtained by combining the two dimensions of richness and balance. Compared with the method in the related art that can only reflect diversity from one dimension of richness or balance, the content diversity determined by this application is more comprehensive and more accurate.

[0097] Regarding richness, the related art measures richness by the number of exposed categories, where the number of exposed categories can refer to the number of non-repeated categories that appear in the exposed content. For example, if N content is exposed, the number of non-repeated categories that appear can reflect the number of categories corresponding to the N content, so it can be used to measure richness. In actual applications, the fewer categories a user sees, the more difficult it is to discover new interests, resulting in information limitations.

[0098] Regarding the degree of balance, the related technology uses the aforementioned screen dominance index or cocoon index to preliminarily measure the degree of balance. In actual applications, the screen dominance index and the cocoon index embody similar concepts, and are both used to reflect the proportion of similar content in N content. That is to say, the related technology uses the proportion of top categories in the content set to reflect the degree of balance. For example, N=10, where there are 6 top categories of content, then the value of the screen dominance index or the cocoon index is 0.6, indicating that the difference between the number of content in different categories among the 10 contents is large, and it can be considered that the degree of balance is low. The difference between the screen dominance index and the cocoon index is usually the difference in the time dimension. In practical applications, the screen dominance index is often used to observe the proportion of similar content in the content exposed in a short period of time, and the cocoon index is often used to observe the proportion of similar content in the content exposed in a longer period of time. For example, a typical definition of the cocoon index refers to the proportion of similar content in the content exposed to users within 3 days (if it is greater than 30%, it can be considered that an information cocoon has been formed). It can be seen that the difference between the screen dominance index and the cocoon index is only the difference in time range. After aligning the time dimension, both are used to reflect the proportion of similar content in the exposed content. The reason for distinguishing them as two indicators is that the two different time scales correspond to different user experiences. The repetitive experience brought to users after similar content is exposed in a short period of time, that is, in a smaller time scale. For example, when users see similar content on one screen, the repetitive experience is stronger than when they see similar content the next day. By defining two indicators with different time scales, it can be flexibly applied to different scenarios.

[0099] It should be noted that although the proportion of top categories is widely used as a preliminary measure of balance because of its simple calculation and intuitive meaning, it is difficult to accurately measure the balance by the proportion of top categories. For this reason, the relevant technology also provides a way to use the Gini coefficient to more completely measure the balance. In addition to considering the proportion of top categories, the Gini coefficient also considers the proportion of other categories, so it can more completely measure the balance. In practical applications, the Gini coefficient was originally a scalar measure of the degree of disparity between the rich and the poor. The original definition refers to collecting the total wealth of everyone in society, sorting them from small to large, calculating the cumulative function, and then drawing the following Figure 6 The Lorenz Curve shown in the figure shows the cumulative distribution of population ratios on the horizontal axis and the cumulative distribution of wealth ratios on the vertical axis. When wealth distribution is balanced, the Lorenz Curve is close to the 45-degree line. When wealth distribution is unbalanced, the Lorenz Curve is far away from the 45-degree line. Therefore, A can be used to represent the area above the Lorenz Curve, B to represent the area below the Lorenz Curve, and A / (A+B) to represent the degree of balance. When the Gini coefficient is used to measure the degree of balance in the recommendation scenario, Figure 6In the Lorenz curve shown in , each w can represent a category, and the value of each w on the vertical axis can be considered as the proportion of the number of contents under this category. It is generally believed that the higher the Gini coefficient, the worse the balance. In practical applications, the Gini coefficient can be determined by the following formula:

[0100]

[0101] In the above formula, G can represent the value of the Gini coefficient, A can represent the area above the Lorenz curve, and B can represent the area under the Lorenz curve.

[0102] Among them, the area A on the Lorentz curve can be determined by the following formula:

[0103]

[0104] In the above formula, m can represent the number of species, w i It can represent the proportion of content in the i-th category, that is, the exposure rate of the i-th category.

[0105] And, the area B on the Lorentz curve can be determined by the following formula:

[0106]

[0107] After determining A and B, the corresponding Gini coefficient can be determined by the following formula:

[0108]

[0109] In practical applications, the smaller the Gini coefficient, the higher the balance, so 1-G can usually be used directly as an indicator to measure balance. However, although the Gini coefficient can solve the shortcomings of the screen dominance index and the cocoon index, the Gini coefficient can only reflect content diversity from the dimension of balance, and it is still difficult to accurately quantify content diversity. For example, N = 6, m = 2 for the first content set, and the number of contents in each category is 3, and m = 3 for the second content set, and the number of contents in each category is 2. The Gini coefficients of the two are the same, but from the actual recommendation effect and user experience, the latter can show users more diverse content. It can be seen that the Gini coefficient cannot accurately measure content diversity.

[0110] It can be seen that the various indicators used in related technologies can only measure content diversity from one dimension, richness or balance, making it difficult to accurately quantify content diversity. If you want to cover these two dimensions, you need to fuse at least two indicators. When fusion, weights are usually set artificially to weight the values ​​of at least two indicators. However, this will bring subjective bias, resulting in poor accuracy of content diversity.

[0111] Compared with the related art, after adopting the diversity determination method provided by the present application, the two dimensions of richness and balance can be integrated to obtain an overall result. In practical applications, it can be considered that the diversity determination method provided by the present application is a composite single indicator that integrates the two dimensions of richness and balance, that is, no longer focusing on the results of richness and balance alone, but integrating these two dimensions to obtain an overall result, which can more comprehensively reflect the diversity of content, and no longer need to set weights, etc., thereby avoiding subjective bias and improving the accuracy of content diversity. For a better understanding, the embodiment of the present application provides an example of the implementation method of constructing a composite single indicator covering the two dimensions of richness and balance:

[0112] In practical applications, the classification repetition rate may refer to the probability that two items are not of the same kind when two items are randomly selected from a set including multiple items. It can be understood that the greater the probability that the two items are not of the same kind, the higher the diversity of the set. For a content set that recommends N contents, the classification repetition rate can be understood as the probability of whether the content that a user has seen is repeated when seeing a content. If it is repeated, it indicates that the content diversity is poor. If it is not repeated, it indicates that the user has seen new content, and the content diversity is good. Therefore, in this application, the classification repetition rate can be selected to construct a composite single indicator provided by this application. Specifically, the classification repetition rate indicator can be proposed in two ways, including sampling with replacement and sampling without replacement, both of which are more commonly used. Among them, sampling without replacement means taking one from N contents for the first time and not putting it back after taking it out. Sampling with replacement means taking one from N contents for the first time and putting it back after taking it out. That is, the difference is whether the second sampling is sampled from N-1 or N. It can be understood that the sum of the probability that two contents do not belong to the same category and the probability that two contents belong to the same category is equal to 1. Therefore, based on the original definition of classification repetition rate "the probability that two randomly selected contents do not belong to the same category", the composite single indicator provided by this application can be determined by the following formula:

[0113] S=P(x1≠x2)=1-P(x1=x2)

[0114] In the above formula, S can represent the value of a composite single indicator, P can represent the probability, x1 can represent the category of the content extracted for the first time, and x2 can represent the category of the content extracted for the second time.

[0115] In practical applications, when the number of categories corresponding to N content is m, it can also be considered that the content set is divided into m subsets, and the content in each subset is the content belonging to the same category. On this basis, the repetition probability of each subset can be calculated first and then the sum of the m subsets can be calculated. Correspondingly, the composite single index can be determined by the following formula:

[0116]

[0117] In the above formula, S can represent the value of a composite single indicator, i can represent the i-th subset, and m i can represent the category of the i-th subset, It can represent the probability that the categories of the two contents are both the categories of the i-th subset.

[0118] In order to simplify the description, the present application takes the sampling method with replacement as an example. In the sampling method with replacement, the two samplings do not affect each other, that is, the probability that the two contents of the two samples hit the same category can be expressed as the product of the probability of each sampling hitting the category. Correspondingly, the composite single index can be determined by the following formula:

[0119]

[0120] In the above formula, S can represent the value of a composite single indicator, N can represent the number of contents in the content set, and n i The number of contents in the i-th subset can be represented, that is, the number of contents in the i-th category can be represented.

[0121] It can be understood that for the sampling without replacement method, since the second sampling is sampled from N-1 contents and the number of contents under the i-th category has been reduced by one, the probability that the category of the content sampled for the second time hits the i-th category can be expressed as Correspondingly, under the sampling method without replacement, the composite single index can be determined by the following formula:

[0122]

[0123] Through the above derivation, the formula for determining the composite single index can be preliminarily determined. For better understanding, the present embodiment still takes the sampling method with replacement as an example to derive the composite single index shown in the above formula to comprehensively cover the richness and balance:

[0124] First, regarding the richness, we can assume that the balance of the content set is 1, that is, the number of contents in each category is the same, that is, Correspondingly, the composite single indicator can be determined by the following formula:

[0125]

[0126] Based on this, it can be seen that when the balance degree is equal to 1, the value S of the composite single indicator is positively correlated with the number of species m. Extending it to other balance degrees, it can still be shown that the results of the composite single indicator are related to richness.

[0127] Similarly, for the degree of balance, we can assume that the number of types m = 2, then n1 = N-n2. Correspondingly, the composite single indicator can be determined by the following formula:

[0128]

[0129] Then, by taking the derivative of n1, we can get:

[0130]

[0131] It can be seen that the derivative decreases with respect to n1, so the original function (that is, the expression of the composite single index) is a convex function. When the derivative is zero, the original function takes the maximum value. Correspondingly, It can be determined that n1=n2. At this time, the number of contents under the two categories is the same, indicating that the balance is optimal, thus proving that the result of the composite single indicator is related to the balance. Extending to the case where m is greater than 2, it can still be shown that the result of the composite single indicator is related to the richness.

[0132] It can be proved through the above derivation that when the balance is the same, the value of the composite single indicator is positively correlated with the richness, and when the richness is equivalent, the value of the composite single indicator is positively correlated with the balance. Regardless of how the richness and balance change, the value of the composite single indicator is a result of the integrity determined based on these two dimensions, so it can cover both the richness and balance dimensions at the same time. Compared with the various indicators used in the relevant technology, it is more comprehensive and can avoid the deviation caused by setting weights, so it can more accurately reflect the diversity of content. Therefore, it can be considered that the composite single indicator provided in this application can cover the functions of other indicators used in the relevant technology, and achieve a more comprehensive and accurate effect.

[0133] In order to better understand the effect of the composite single index provided by the present application, the present application takes the evaluation of the content diversity of five given content sets of N=4 as an example to compare and illustrate the composite single index provided by the present application and the index provided in the related art. Among them, the given five content sets can be<A,A,A,A> ,<A,A,A,B> ,<A,A,B,B> ,<A,A,B,C> as well as<A,B,C,D> , if the related technology uses the exposure classification number to determine the richness and the Gini coefficient to determine the balance, then for these five content sets, the results shown in the following table can be obtained:

[0134]

[0135]

[0136] It can be seen that the indicators provided in the relevant technology can only reflect the content diversity from one dimension of richness or balance, which is relatively one-sided, while the composite single indicator provided by this application can cover both dimensions and is more comprehensive. At the same time, in terms of the interpretation of content diversity, after adopting the richness and balance determined in the relevant technology, for these five content sets, we can draw a graph with balance as the horizontal axis and richness as the vertical axis. Figure 7 The distribution diagram shown in the figure can more intuitively explain the content diversity of each content combination from the changes in these two dimensions.<A,A,A,A> become<A,A,A,B> , which can be explained as seeing a new category B, so the content diversity is higher, that is, the content diversity is improved due to the increase in richness. Similarly, the content collection is changed from<A,A,A,B> become<A,A,B,B> , the richness remains the same, and the content diversity is improved due to the improvement of balance, and the content collection<A,B,C,D> It can be considered that both the richness and balance have reached the optimal solution, with the highest content diversity. However, in this interpretation, the content set<A,A,B,B> become<A,A,B,C> In the case of , since the richness increases and the balance decreases, it is difficult to explain the difference in content diversity between the two content sets. This fully illustrates that the indicators used in the relevant technology are difficult to apply to various situations. If the values ​​of the two dimensions of richness and balance are weighted and fused, it will involve setting weights, which will bring about subjective bias and still be difficult to accurately reflect the content diversity. In this regard, after adopting the composite single indicator provided by this application, we no longer need to pay attention to the results of a certain dimension, but instead integrate the two dimensions to obtain a holistic result, which is suitable for the explanation of content diversity in various situations and is more applicable. For example, the content set is from<A,A,B,B> become<A,A,B,C> In the case of the latter, although the richness is improved and the balance is reduced, after combining these two dimensions, it can be determined that the content diversity of the latter is higher than that of the former.

[0137] Through the above embodiments, the composite single indicator of the present application that integrates the two dimensions of richness and balance is illustrated. It can be understood that in one possible implementation method, the composite single indicator can be used alone, and the result obtained can refer to the aforementioned initial diversity. On this basis, it is also possible to comprehensively consider whether there is a situation where content of the same category is densely arranged and make corrections to further improve the accuracy of content diversity. In another possible implementation method, it is also possible to combine the diversity factors of the aforementioned target content combination, based on which, it is possible to simultaneously cover the richness, balance, and whether similar content is densely arranged to determine the overall result, so that content diversity can be quantified more accurately. For a better understanding, the embodiment of the present application still takes the aforementioned sampling method without replacement as an example, and provides the following example description:

[0138] In the sampling without replacement method, the classical probability model can be used to solve the probability that the two contents obtained by two samples belong to the same category. Specifically, it can be done by exhaustively enumerating possible sampling events and then counting the number of equal events. In this way, the probability that the two sampled contents belong to the same category can be determined by judging whether the categories of the i-th content and the j-th content are the same, and it is no longer necessary to rely on the m divided subsets. Correspondingly, the aforementioned composite single indicator can be determined by the following formula:

[0139]

[0140] In the above formula, x i Can represent the category of the i-th content, x j It can represent the category of the jth content, Can be an indicative function, used to represent x i and x j Whether the corresponding values ​​are equal.

[0141] In order to simplify calculations and save computing resources, in the process of exhaustive combinations, it can be considered that sampling first to obtain x1 and then sampling to obtain x2 is equivalent to sampling first to obtain x2 and then sampling to obtain x1. Therefore, it can be considered that the content combination consisting of the i-th content and the j-th content, and the content combination consisting of the j-th content and the i-th content are equivalent. Based on this, the number of exhaustive combinations can be reduced by half. Accordingly, the above formula can be further simplified to:

[0142]

[0143] In the above formula, It can represent the number of content combinations formed by combining the i-th content and the j-th content among the N contents, that is, it can represent the aforementioned R.

[0144] Based on the above derivation, it can be shown that the above formula can still cover the two dimensions of richness and balance. As an indicative function, it represents x i and x jWhether the corresponding values ​​are equal or not can reflect the similarity between the i-th content and the j-th content, but there are only two values ​​of 0 and 1. In this way, the classification of the content will directly affect the final content diversity. It can be understood that for a given N content, when determining the category of each content, two contents may be classified as the same category when there are fewer optional categories (optional categories can be manually specified), and may be classified as different categories when there are more optional categories, that is, the same two contents may belong to the same category or different categories under different classification systems, but their original content similarity is unique, but they may be determined to be the same category or different categories according to different thresholds or different numbers of optional categories. In practical applications, the category of content can usually be determined by a clustering algorithm or a classification algorithm, which can essentially be considered as the result of calculating the content similarity based on the feature vector of the content. In other words, although the category determined based on the classification will be affected by the number of optional categories, the content similarity between the two contents is not affected, for example, the content similarity determined based on the cosine similarity is not affected.

[0145] Therefore, in order to reduce the interference and subjective bias caused by manually specifying optional categories, and to more accurately quantify the similarity between the ith content and the jth content, the indicative function can be expanded to the aforementioned content similarity, based on which it is helpful to more accurately quantify the similarity between the ith content and the jth content. At the same time, in order to be able to integrate the aforementioned diversity factors and thus consider the impact of whether similar content is densely arranged on content diversity, the aforementioned expression of Y and the cosine similarity between the feature vectors of two contents can be used as an example to represent the content similarity. Based on this, the above formula can be further improved to:

[0146]

[0147] Based on this, not only can the two dimensions of richness and balance be covered at the same time, but also the content similarity and distance factor of each content combination can be considered at the same time, integrating the impact of all dimensions on content diversity, which is conducive to more accurate quantification of content diversity. Based on the above derivation, it can be proved that the above formula can cover richness, balance content similarity and distance factor at the same time, and the final value of S can reflect the content diversity. Therefore, in one possible implementation method, the above formula can be directly used to quantify the content diversity of the content set, which is simpler and more convenient, and is conducive to improving efficiency.

[0148] Based on this, it can be considered that the embodiment of the present application provides a content diversity measurement index covering various dimensions, and this measurement index is a single index, which is more concise and can also avoid the subjective bias caused by setting parameters and weights in related technologies. At the same time, covering multiple dimensions can quantify content diversity more comprehensively and accurately. In other words, the use of this application can unify the measurement method of content diversity under an overall theoretical framework, solve the problem of imperfect definition and determination method of content diversity in practical applications, and derive and sort out the relationship between multiple dimensions such as richness, balance, content similarity and distance factor, which is helpful for content diversity analysis in actual business. Among them, the comprehensive distance factor and content similarity can well measure whether similar content is densely arranged, that is, it can well measure the exposure density of similar content. For content sets that are more dense in order, lower content diversity will be determined, which is conducive to avoiding dense exposure when guiding recommendations, reducing the situation of users seeing similar content in succession, and avoiding giving users a repeated experience. Therefore, it can be considered that the content diversity measurement index provided by this application contains information in the time dimension, which is one of the important dimensions when determining content diversity, and is the third dimension orthogonal to richness and balance. Based on the content diversity measurement indicators provided in this application, a single indicator can be used to cover richness, balance and time dimensions, and content diversity can be quantified more comprehensively and accurately.

[0149] It should be noted that the above examples are constructed based on the classification repetition rate. In practical applications, the classification repetition rate can also be called the Simpson index, so it can also be considered that the content diversity measurement index provided by this application can be constructed based on the Simpson index. It is understandable that in addition to the above classification repetition rate, it can also be constructed based on indicators such as information entropy. The derivation process is similar to the above process and will not be repeated here.

[0150] Through the above embodiments, the diversity determination method provided by the present application is described in detail. For better understanding, the present application also provides an embodiment of the application stage of the diversity determination method, which is as follows:

[0151] In the actual recommendation scenario, the product developers may design a recommendation strategy, and then determine the content set based on the designed recommendation strategy to achieve the purpose of recommending the content set to users through the product. It is understandable that the content diversity of the content set determined by different recommendation strategies is different, which will affect the recommendation effect that can be obtained after the recommendation. Therefore, in a possible implementation, the diversity determination method provided by the present application can be applied to the internal analysis and iteration process of the product, mainly for the product developers. In the specific implementation, the recommendation result obtained by recommending the content set in the target product can be obtained, wherein the target product can refer to the product that recommends the content set to the user, for example, it can be the aforementioned information flow product, the content set can be generated based on the recommendation strategy for the target product, and the recommendation strategy can refer to the new strategy designed by the developer of the target product, which is used to indicate the generation of the content set so as to recommend it to the user through the target product. The recommendation result can reflect the recommendation effect after the content set is recommended, such as the feedback of the user after the content set is recommended to the user, etc. In actual applications, the recommendation result can characterize the advantages and disadvantages of the recommendation strategy. If the recommendation result indicates that a good recommendation effect is obtained, it is considered that the recommendation strategy is good enough, otherwise the recommendation strategy is not good enough. Correspondingly, if the recommendation result indicates that the recommendation strategy meets the adjustment conditions, it means that the recommendation strategy is not good enough and needs to be adjusted and optimized. Since the recommendation effect caused by different content diversity will be different, when the recommendation effect is poor, the reason why the recommendation strategy is difficult to obtain a good recommendation effect can be explained by quantifying the content diversity. For example, the poor recommendation effect is caused by the low content diversity of the content set determined under this recommendation strategy. Therefore, the strategy adjustment target can be determined based on the content diversity of the content set. The strategy adjustment target can be used to indicate the direction and target when adjusting the recommendation strategy. Finally, the pending recommendation strategy can be adjusted according to the strategy adjustment target, so that the adjusted recommendation strategy can obtain a good recommendation effect. Based on this, by quantifying the content diversity, the optimization of the recommendation strategy can be guided in the analysis and iteration process of the product. By assisting R&D personnel to evaluate the recommendation strategy of the target product more objectively and scientifically, it is conducive to R&D personnel to iterate the recommendation strategy that can obtain a good recommendation effect more quickly, realize the iterative upgrade of the target product recommendation system, and improve the user experience and recommendation benefits after the recommendation strategy should be good.

[0152] In practical applications, the internal analysis and iteration process of the target product can be found in Figure 8As shown, specifically, R&D personnel can design new recommendation strategies. For new recommendation strategies, they generally observe the user feedback after the application of the new recommendation strategies through grayscale experiments. That is, first, grayscale experiments can be conducted, and the experimental data of grayscale experiments can be collected to determine the recommendation results. Then, the recommendation strategies can be evaluated and analyzed to complete the effect analysis. Finally, based on the effect analysis, it can be determined whether to adjust the recommendation strategies to achieve strategy iteration. Usually, the content diversity determination method provided in this application can be applied to the effect analysis link, that is, the quantitative content diversity is used to explain the reasons why the recommendation strategy obtains such a recommendation effect. In actual applications, if the recommendation effect is determined to be good based on the effect analysis, the recommendation strategy is considered to be good enough, so the recommendation strategy can be iterated into the strategy of the target product. If the recommendation effect is determined to be poor based on the effect analysis, the recommendation strategy is considered to be not good enough. At this time, the recommendation strategy needs to be adjusted to achieve strategy iteration. In practical applications, the higher the content diversity of the content set determined based on the recommendation strategy, the more dissimilar content can be recommended to users. That is, by recommending different types of content, the recommendation system of the target product can better meet the diverse interests of users and provide users with richer content choices. From the user's perspective, recommending more dissimilar content can bring more surprises and freshness, which is conducive to improving user experience.

[0153] In practical applications, user interests and needs are diverse, and different users have different needs for recommended content sets, which is mainly reflected in the different preferences of different users for the content diversity of the obtained content sets. For example, some users prefer to watch content that they are already familiar with or concerned about, and some users prefer to watch some content that they have never been exposed to, to help themselves expand their areas of interest, etc. Therefore, in another possible implementation, the content diversity determination method provided by the present application can also be applied to the stage of recommending content sets to users, by combining the user's diversity preference conditions, in order to achieve personalized recommendations. In specific implementation, users who need to recommend content sets can be used as target users, and accordingly, the diversity preference conditions of target users can be used to characterize the diversity threshold of the target user for the recommended content set (for example, the diversity threshold preferred by the target user is [0.6, 0.8]). Accordingly, if it is determined that the content diversity of the content set meets the diversity preference conditions of the target user, it means that the value of the content diversity meets the diversity threshold of the target user, and a good recommendation effect can be obtained after recommendation, so the content set can be recommended to the target user. Correspondingly, if it is determined that the content diversity of the content set does not meet the diversity preference condition, it means that the value of the content diversity does not meet the diversity threshold of the target user, and it is difficult to obtain a good recommendation effect after recommendation. For the same N content, different content sorting will affect the value of content diversity. Therefore, at this time, the content sorting can be adjusted according to the diversity preference condition to obtain the target set. Correspondingly, if a target set whose content diversity meets the diversity preference condition is obtained, it means that the content diversity of the N content can be made to meet the diversity threshold of the target user by adjusting the content sorting, so the target set can be recommended to the target user. Based on this, by applying the content diversity determination method provided by the present application to the stage of recommending content sets to users, and combining the user's diversity preference conditions, personalized recommendations can be achieved, which is conducive to improving the recommendation effect.

[0154] It can be seen that the present application also provides a recommendation method based on content diversity, which can be a method of adjusting content sorting based on diversity preference conditions, so that the same N content can meet the purpose of different users' diversity preference conditions, meet the different preferences and needs of different users, and help improve the recommendation effect and user experience, and encourage users to use the target product for a longer period of time. For a better understanding, please refer to Fig. 9 As shown, Fig. 9 The diversity threshold of the target user is [0.6, 0.8], indicating that the target user is Fig. 9The content diversity distribution of the preferred content set in the information flow product shown. Accordingly, when recommending a content set to the target user, taking N=4 as an example, if the content set is determined to be an ordered set arranged according to the first content, the second content, the third content, and the fourth content based on the content sorting, the content similarity between the first content and the second content is 0.9, which is very similar content, and the content similarity between any other two contents is assumed to be 0.1, which can be considered to be not very similar content. In this way, the content diversity of the content set is 0.3, and the reason is that the similar first content and the second content appear densely arranged. Since this target user prefers a more diverse content set, direct recommendation will give the target user a sense of repetition of seeing similar content in succession, which is a poor experience. At this time, the content sorting of the content set can be adjusted based on the diversity threshold of [0.6, 0.8]. Specifically, 0.3 can be adjusted toward [0.6, 0.8]. Finally, after the adjustment, the target set is arranged according to the first content, the third content, the fourth content, and the second content, and its content diversity is 0.7. Since the similar first content and second content are arranged at intervals, the exposure density of similar content is reduced, thereby improving the content diversity and meeting the diversity preference of the target user. At this time, the target set can be recommended to the target user. Fig. 9 An example of recommending a content set to a user based on an information flow product is shown in FIG. 9 , wherein a computer 900 may run an information flow product, such as a smart phone, and the information flow product is logged in to the target user's account. When the information flow product is running, Fig. 9 The user interface shown in the figure can display these four contents in the recommended target set according to the order of the contents of the target set. The display effect is as follows Fig. 9 shown.

[0155] It should be noted that the present application does not impose any limitation on how to determine the diversity preference condition. For ease of understanding, the present application embodiment provides the following example, that is, the diversity preference condition of the target user can be determined in the following manner:

[0156] First, multiple historical collections recommended to the target user can be obtained, as well as the target user's interaction data for the multiple historical collections. Among them, the multiple historical collections may refer to content collections that have been recommended to the target user, and each of the multiple historical collections includes at least three contents, so the content diversity of each historical collection can be more accurately quantified based on the diversity determination method provided in the present application. In actual applications, the target user's interaction data for the historical collection can reflect the user's preference for the content collection, such as the interaction data may include the target user's likes, comments, viewing time, browsing time and other data for the content in the historical collection. Then, the diversity preference condition can be determined based on the interaction data and the content diversity corresponding to the multiple historical collections. Since the interaction data can reflect the user's preferences, it is helpful to determine accurate diversity preference conditions.

[0157] It should be noted that, based on the implementation methods provided in the above aspects, this application can also be further combined to provide more implementation methods.

[0158] based on Figure 3 Corresponding to the diversity determination method provided in the embodiment, the embodiment of the present application further provides a diversity determination device 1000, and the diversity determination device 1000 may include an acquisition unit 1001, a combination unit 1002 and a determination unit 1003:

[0159] The acquisition unit 1001 is used to acquire a content set including N contents, and acquire a content ranking of the content set; the content ranking is used to represent the display order of the N contents when recommending the content set, and N is an integer greater than 2;

[0160] The combining unit 1002 is used to combine the i-th content and the i+1-th content that are adjacent to each other in the content sorting of the N contents to obtain M first content combinations corresponding to the content set; i is an integer greater than or equal to 1 and less than N;

[0161] The determining unit 1003 is configured to determine the content diversity of the content set according to the content similarities respectively corresponding to the M first content combinations; the content similarity of the first content combination is determined based on the similarity between two contents in the first content combination.

[0162] In a possible implementation, the combination unit is further used for:

[0163] Combining the i-th content and the j-th content of the N contents based on the content sorting to obtain K second content combinations corresponding to the content set; j is an integer greater than i+1 and less than or equal to N;

[0164] The determining unit is further configured to:

[0165] The content diversity of the content set is determined according to the content similarities corresponding to the R target content combinations respectively; the R target content combinations include M first content combinations and K second content combinations, and the content similarity of the second content combination is determined based on the similarity between two contents in the second content combination.

[0166] In a possible implementation manner, the determining unit is further configured to:

[0167] determining a distance factor of the target content combination based on a ranking difference between two contents in the target content combination in the content ranking;

[0168] The content diversity of the content set is determined according to the content similarities and distance factors respectively corresponding to the R target content combinations.

[0169] In a possible implementation manner, the determining unit is further configured to:

[0170] determining a diversity factor of the target content combination with respect to the content diversity according to the content similarity of the target content combination and the distance factor of the target content combination;

[0171] The content diversity of the content set is determined based on the diversity factors corresponding to the R target content combinations.

[0172] In a possible implementation manner, if the M first content combinations are all first content combinations in the content set, and if the content similarity of the first content combinations is determined based on the content similarity between two contents in the first content combinations, the determining unit is further configured to:

[0173] Determine N categories corresponding to the N contents according to the content similarities corresponding to the M first content combinations, and count the number of categories corresponding to the N categories; arrange the N categories according to a category ranking, and the category ranking is consistent with the content ranking;

[0174] If the number of categories is greater than or equal to 2 and less than N, the content diversity of the content set is determined according to the number of contents in the content set, the number of categories, the number of contents under each category, and the category ranking.

[0175] In a possible implementation manner, the determining unit is further configured to:

[0176] If the number of categories is equal to 1 or the number of categories is equal to N, the content diversity of the content set is determined according to the number of contents in the content set, the number of categories, the number of contents under each category, and the content similarities corresponding to the M first content combinations.

[0177] In a possible implementation manner, the determining unit is further configured to:

[0178] determining an initial diversity of the content set according to the number of contents in the content set, the number of categories, and the number of contents under each category;

[0179] If it is determined based on the category sorting that categories of the same type in the N categories are arranged adjacently, the initial diversity is downwardly modified to obtain the content diversity of the content set.

[0180] In a possible implementation, the device further includes a recommendation unit and an adjustment unit:

[0181] The recommendation unit is configured to recommend the content set to the target user if it is determined that the content diversity of the content set meets the diversity preference condition of the target user; the diversity preference condition is used to represent the diversity threshold of the target user for the recommended content set;

[0182] The adjusting unit is configured to adjust the content sorting according to the diversity preference condition to obtain a target set if it is determined that the content diversity of the content set does not satisfy the diversity preference condition;

[0183] The recommendation unit is further configured to recommend the target set to the target user if a target set whose content diversity satisfies the diversity preference condition is obtained.

[0184] In a possible implementation, the diversity preference condition is determined in the following manner:

[0185] Acquire multiple history sets recommended to the target user, and acquire interaction data of the target user for the multiple history sets; each of the multiple history sets includes at least three contents;

[0186] The diversity preference condition is determined according to the content diversity corresponding to the interaction data and the multiple history sets respectively.

[0187] In a possible implementation manner, the acquiring unit is further configured to:

[0188] Acquire a recommendation result obtained by recommending the content set in a target product; the content set is generated based on a recommendation strategy for the target product;

[0189] If the recommendation result indicates that the recommendation strategy meets the adjustment condition, determining a strategy adjustment target based on the content diversity of the content set;

[0190] The pending recommendation strategy is adjusted according to the strategy adjustment target.

[0191] It can be seen from the above technical solution that first, a content set including N contents can be obtained, and the content ranking of the content set can be obtained, where N is an integer greater than 2. Among them, the content ranking can be used to characterize the display order of N contents when recommending the content set, so the N contents can be combined based on the adjacent i-th content and i+1-th content based on the content ranking to obtain M first content combinations corresponding to the content set, where i is an integer greater than or equal to 1 and less than N. Based on this, it can be determined what the closely arranged contents are among the N contents. Finally, the content diversity of the content set can be determined based on the content similarities corresponding to the M first content combinations. Among them, the content similarity of the first content combination can be determined based on the similarity between two contents in the first content combination, which can reflect the similarity between the two contents, and since the display order of the two contents is adjacent, the content similarity based on the first content combination can measure whether similar contents are densely arranged, and whether similar contents are densely arranged will affect the recommendation effect of the content set. Based on this, the content diversity can reflect whether similar contents are densely arranged, so that the content diversity can be quantified more accurately.

[0192] The embodiment of the present application further provides a computer device, which may be a terminal. For example, the terminal is a smart phone:

[0193] Fig.11 The block diagram shows a partial structure of a smart phone provided in an embodiment of the present application. Fig.11 The smartphone includes: a radio frequency (RF) circuit 1110, a memory 1120, an input unit 1130, a display unit 1140, a sensor 1150, an audio circuit 1160, a wireless fidelity (WiFi) module 1170, a processor 1180, and a power supply 1190. The input unit 1130 may include a touch panel 1131 and other input devices 1132, the display unit 1140 may include a display panel 1141, and the audio circuit 1160 may include a speaker 1161 and a microphone 1162. Those skilled in the art will appreciate that Fig.11 The structure of the smartphone shown in the figure does not constitute a limitation of the smartphone, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0194] The memory 1120 can be used to store software programs and modules. The processor 1180 executes various functional applications and data processing of the smartphone by running the software programs and modules stored in the memory 1120. The memory 1120 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the smartphone (such as audio data, a phone book, etc.), etc. In addition, the memory 1120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0195] The processor 1180 is the control center of the smartphone, which uses various interfaces and lines to connect various parts of the entire smartphone, and executes various functions of the smartphone and processes data by running or executing software programs and / or modules stored in the memory 1120, and calling data stored in the memory 1120. Optionally, the processor 1180 may include one or more processing units; preferably, the processor 1180 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 1180.

[0196] In this embodiment, the steps performed by the processor 1180 in the smartphone may be based on Fig.11 The structure shown is implemented.

[0197] The computer device provided in the embodiment of the present application may also be a server. Fig.12 As shown, Fig.12 The structural diagram of the server 1200 provided in the embodiment of the present application, the server 1200 may have relatively large differences due to different configurations or performances, and may include one or more processors, such as a central processing unit (CPU) 1222, and a memory 1232, one or more storage media 1230 (such as one or more mass storage devices) storing application programs 1242 or data 1244. Among them, the memory 1232 and the storage medium 1230 can be short-term storage or permanent storage. The program stored in the storage medium 1230 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1222 can be configured to communicate with the storage medium 1230 and execute a series of instruction operations in the storage medium 1230 on the server 1200.

[0198] The server 1200 may also include one or more power supplies 1226, one or more wired or wireless network interfaces 1250, one or more input and output interfaces 1258, and / or one or more operating systems 1241, such as Windows Server 2000. TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TM etc.

[0199] In this embodiment, the central processor 1222 in the server 1200 may perform the following steps:

[0200] Acquire a content set including N contents, and acquire a content ranking of the content set; the content ranking is used to represent the display order of the N contents when recommending the content set, and N is an integer greater than 2;

[0201] Combining the i-th content and the i+1-th content that are adjacent to each other based on the content sorting of the N contents to obtain M first content combinations corresponding to the content set; i is an integer greater than or equal to 1 and less than N;

[0202] The content diversity of the content set is determined according to the content similarities respectively corresponding to the M first content combinations; the content similarity of the first content combination is determined based on the similarity between two contents in the first content combination.

[0203] According to one aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium is used to store a computer program. When the computer program is executed by a computer device, the computer device executes the diversity determination method described in the above-mentioned embodiments.

[0204] According to one aspect of the present application, a computer program product is provided, the computer program product comprising a computer program, the computer program being stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the method provided in various optional implementations of the above-mentioned embodiments.

[0205] The descriptions of the processes or structures corresponding to the above-mentioned figures have different emphases. For parts that are not described in detail in a certain process or structure, please refer to the relevant descriptions of other processes or structures.

[0206] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0207] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0208] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0209] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0210] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the relevant technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store program codes.

[0211] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, ordinary technical members in the art should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A diversity determination method, characterized in that: The method comprises: Acquire a content set including N contents, and acquire a content ranking of the content set; the content ranking is used to represent the display order of the N contents when recommending the content set, and N is an integer greater than 2; Combining the i-th content and the i+1-th content that are adjacent to each other based on the content sorting of the N contents to obtain M first content combinations corresponding to the content set; i is an integer greater than or equal to 1 and less than N; The content diversity of the content set is determined according to the content similarities respectively corresponding to the M first content combinations; the content similarity of the first content combination is determined based on the similarity between two contents in the first content combination.

2. The method according to claim 1, characterized in that The method further comprises: Combining the i-th content and the j-th content of the N contents based on the content sorting to obtain K second content combinations corresponding to the content set; j is an integer greater than i+1 and less than or equal to N; The determining the content diversity of the content set according to the content similarities respectively corresponding to the M first content combinations includes: The content diversity of the content set is determined according to the content similarities corresponding to the R target content combinations respectively; the R target content combinations include M first content combinations and K second content combinations, and the content similarity of the second content combination is determined based on the similarity between two contents in the second content combination.

3. The method according to claim 2, characterized in that The method further comprises: determining a distance factor of the target content combination based on a ranking difference between two contents in the target content combination in the content ranking; The determining the content diversity of the content set according to the content similarities respectively corresponding to the R target content combinations includes: The content diversity of the content set is determined according to the content similarities and distance factors respectively corresponding to the R target content combinations.

4. The method according to claim 3, characterized in that The determining the content diversity of the content set according to the content similarities and distance factors respectively corresponding to the R target content combinations includes: determining a diversity factor of the target content combination with respect to the content diversity according to the content similarity of the target content combination and the distance factor of the target content combination; The content diversity of the content set is determined based on the diversity factors corresponding to the R target content combinations respectively.

5. The method according to claim 1, characterized in that If the M first content combinations are all first content combinations of the content set, and if the content similarity of the first content combinations is determined based on the content similarity between two contents in the first content combinations, determining the content diversity of the content set according to the content similarities respectively corresponding to the M first content combinations includes: Determine N categories corresponding to the N contents according to the content similarities corresponding to the M first content combinations, and count the number of categories corresponding to the N categories; arrange the N categories according to a category ranking, and the category ranking is consistent with the content ranking; If the number of categories is greater than or equal to 2 and less than N, the content diversity of the content set is determined according to the number of contents in the content set, the number of categories, the number of contents under each category, and the category ranking.

6. The method according to claim 5, characterized in that The method further comprises: If the number of categories is equal to 1 or the number of categories is equal to N, the content diversity of the content set is determined according to the number of contents in the content set, the number of categories, the number of contents under each category, and the content similarities corresponding to the M first content combinations.

7. The method according to claim 5, characterized in that The determining the content diversity of the content set according to the number of contents in the content set, the number of categories, the number of contents under each category, and the category ranking includes: determining an initial diversity of the content set according to the number of contents in the content set, the number of categories, and the number of contents under each category; If it is determined based on the category sorting that categories of the same type in the N categories are arranged adjacently, the initial diversity is downwardly modified to obtain the content diversity of the content set.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: If it is determined that the content diversity of the content set meets the diversity preference condition of the target user, recommending the content set to the target user; the diversity preference condition is used to represent the diversity threshold of the target user for the recommended content set; If it is determined that the content diversity of the content set does not satisfy the diversity preference condition, adjusting the content sorting according to the diversity preference condition to obtain a target set; If a target set whose content diversity satisfies the diversity preference condition is obtained, the target set is recommended to the target user.

9. The method according to claim 8, characterized in that The diversity preference condition is determined in the following manner: Acquire multiple history sets recommended to the target user, and acquire interaction data of the target user for the multiple history sets; each of the multiple history sets includes at least three contents; The diversity preference condition is determined according to the content diversity corresponding to the interaction data and the multiple history sets respectively.

10. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: Acquire a recommendation result obtained by recommending the content set in a target product; the content set is generated based on a recommendation strategy for the target product; If the recommendation result indicates that the recommendation strategy meets the adjustment condition, determining a strategy adjustment target based on the content diversity of the content set; The pending recommendation strategy is adjusted according to the strategy adjustment target.

11. A diversity determination device, characterized in that: The device comprises an acquisition unit, a combination unit and a determination unit: The acquisition unit is used to acquire a content set including N contents, and to acquire a content ranking of the content set; the content ranking is used to represent a display order of the N contents when recommending the content set, where N is an integer greater than 2; The combining unit is used to combine the i-th content and the i+1-th content that are adjacent to each other in the content sorting of the N contents to obtain M first content combinations corresponding to the content set; i is an integer greater than or equal to 1 and less than N; The determining unit is configured to determine the content diversity of the content set according to the content similarities respectively corresponding to the M first content combinations; the content similarity of the first content combination is determined based on the similarity between two contents in the first content combination.

12. A computer device, characterized in that: The computer device comprises a processor and a memory: The memory is used to store a computer program and transmit the computer program to the processor; The processor is configured to execute the method according to any one of claims 1 to 10 according to the instructions in the computer program.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and when the computer program is executed by a computer device, the computer device executes the method according to any one of claims 1 to 10.

14. A computer program product comprising a computer program, characterized in that When the method is executed on a computer device, the computer device is enabled to execute the method according to any one of claims 1 to 10.

Images