CN113763027A

CN113763027A - Recommendation information processing method, recommendation information generating method and device

Info

Publication number: CN113763027A
Application number: CN202110503431.XA
Authority: CN
Inventors: 刘凯; 杨秀金
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-05-10
Filing date: 2021-05-10
Publication date: 2021-12-07
Anticipated expiration: 2041-05-10
Also published as: CN113763027B

Abstract

The disclosure provides a recommendation information processing method, a recommendation information generating method and a recommendation information generating device. The recommendation information processing method comprises the following steps: acquiring a recommendation information set containing a plurality of recommendation information; extracting feature elements of a plurality of categories of each piece of recommendation information in the recommendation information set to obtain a feature element set; for each class of feature elements: dividing recommendation information in the recommendation information set into at least two groups, wherein the attribute of the characteristic element of the category of the recommendation information in each group of the at least two groups is mutually exclusive with the attribute of the characteristic element of the category of the recommendation information in other groups of the at least two groups; acquiring at least one result data corresponding to the recommendation information in each group; for each type of result data, a degree of difference between any two groups of result data is calculated, and in the case where the degree of difference is greater than a first predetermined threshold, it is determined that the feature element of the category has significance to the result data.

Description

Recommendation information processing method, recommendation information generating method and device

Technical Field

The present disclosure relates to the multimedia field, and more particularly, to a recommendation information processing method, a recommendation information generating method, and an apparatus.

Background

Advertisements have become an important link in the commodity market, and in order to understand or predict the advertisement delivery effect or meet the demand of producing high-quality advertisements, the advertisements need to be effectively analyzed and evaluated. With the rapid development of internet technology, the number of advertisements is greatly increased, and compared with the traditional advertisements mainly based on image-text advertisements, the video advertisements are more and more popular nowadays, and the receptivity of the audience to the video advertisements is gradually increased. However, most of the conventional advertisement analysis methods are suitable for the image-text advertisements, and cannot distinguish the characteristics of the video advertisements from the image-text advertisements, or cannot be applied to the video advertisements at all, so that it is difficult to accurately evaluate and predict the video advertisements.

Disclosure of Invention

In order to solve the above-described problems, the present disclosure provides a recommended information processing method, a recommended information generation method, a recommended information processing apparatus, a recommended information generation apparatus, a device, a computer-readable storage medium, and a computer program product.

According to an aspect of an embodiment of the present disclosure, there is provided a recommendation information processing method including: acquiring a recommendation information set containing a plurality of recommendation information; extracting feature elements of a plurality of categories of each piece of recommendation information in the recommendation information set to obtain a feature element set; for each class of feature elements in the set of feature elements: dividing a plurality of recommendation information in the recommendation information set into at least two groups, wherein attributes of feature elements of the category of recommendation information within each of the at least two groups are mutually exclusive from attributes of feature elements of the category of recommendation information within other of the at least two groups; acquiring at least one result data corresponding to the recommendation information in each of the at least two groups; for each of the at least one type of result data, a degree of difference between the result data of any two of the at least two groupings is calculated, and in the event that the degree of difference is greater than a first predetermined threshold, it is determined that the feature element of the category is significant to the result data.

According to an example of an embodiment of the present disclosure, the recommendation information processing method further includes: in a case where it is determined that the feature elements of the category have significance to the result data, outputting information about the feature elements of the category, wherein the information includes evaluation information indicating an influence of the feature elements of the category on the result data, the method further includes generating a recommendation information analysis model for providing recommendation information production guidance based on the evaluation information.

According to an example of the embodiment of the present disclosure, wherein the information further includes viewer characteristic information indicating a common characteristic of viewers of recommendation information in a group of which result data is better among the two groups, the recommendation information processing method further includes: and generating recommendation information oriented crowd recommendations based on the viewer characteristic information.

According to an example of an embodiment of the present disclosure, the recommendation information processing method further includes: and predicting the putting effect of the new recommendation information by extracting and analyzing the feature elements of a plurality of categories in the new recommendation information based on the output information about the feature elements of different categories.

According to an example of the embodiment of the present disclosure, the feature elements of the plurality of categories include feature elements of a plurality of basic attribute categories and/or a plurality of content attribute categories, wherein the feature elements of the plurality of basic attribute categories include at least one of a frame rate, a bit rate, a size, a resolution, a saturation, a sharpness, and a number of lenses of recommendation information; the characteristic elements of the plurality of content attribute categories include at least one of a shot, a scene, a person, an object, and a presentation form of the recommendation information.

According to an example of the embodiment of the present disclosure, wherein the at least two groups comprise a first group and a second group, wherein the dividing of the plurality of recommendation information in the set of recommendation information into the at least two groups for each category of feature element in the set of feature elements comprises: for each recommendation information of the plurality of recommendation information, dividing the recommendation information into a first group if a value or a number of feature elements of the category of the recommendation information satisfies a predetermined condition; otherwise, the recommendation information is divided into a second group.

According to an example of the embodiment of the present disclosure, for the feature elements of the plurality of content attribute categories, extracting the feature elements of the plurality of categories of each recommendation information in the recommendation information set includes: extracting feature elements of a plurality of content attribute categories in the recommendation information set using a deep learning model; and for each piece of recommendation information, generating that the recommendation information contains the confidence level of the extracted feature element of each content attribute category and the confidence level of the appearance position of the feature element of the content attribute category in the recommendation information.

According to an example of an embodiment of the present disclosure, the recommendation information processing method further includes: adjusting the output information about the feature elements of different content attribute categories by adjusting the confidence level.

According to an example of an embodiment of the present disclosure, the recommendation information processing method further includes: extracting a timestamp of each feature element in the feature element set; and generating a recommendation information analysis model having time divisions for providing recommendation information production guidance based on the time stamp and the outputted information on the different categories of feature elements, wherein the time stamp is attached to each feature element of the recommendation information at the time of generating each recommendation information of the recommendation information set.

According to an example of the embodiment of the present disclosure, the at least one result data includes at least one part of a launch effect data, a viewer behavior data and an industry distribution data, wherein the launch effect data includes at least one of a recommended information click rate and a recommended information conversion rate, the viewer behavior data includes at least one of a viewing duration, a viewing frequency, a daily repeated viewing frequency, an end-of-play rate, a jump rate, a conversion time point and a conversion success rate of a recommended information viewer, and the industry distribution data indicates a distribution characteristic of an industry to which the recommended information viewer belongs.

According to another aspect of the embodiments of the present disclosure, there is provided a recommendation information generating method, including: extracting characteristic elements of a plurality of categories in the recommendation information to be optimized; for each of the plurality of categories of feature elements: dividing the recommendation information to be optimized into at least two groups, wherein the attribute of the characteristic element of the category of recommendation information in each of the at least two groups is mutually exclusive with the attribute of the characteristic element of the category of recommendation information in the other of the at least two groups; predicting at least one result data of the recommendation information to be optimized according to predetermined evaluation information about the grouping of the characteristic elements aiming at the category based on the grouping where the recommendation information to be optimized is located; optimizing feature elements of the category of the recommendation information to be optimized by using the at least one predicted result data.

According to another aspect of the embodiments of the present disclosure, there is provided a recommendation information processing apparatus including: a recommendation information acquisition unit configured to acquire a recommendation information set containing a plurality of recommendation information; a feature extraction unit configured to extract feature elements of a plurality of categories of each recommendation information in the recommendation information set to obtain a feature element set; and a recommendation information processing unit configured to, for each category of feature elements in the feature element set: dividing a plurality of pieces of recommendation information in the recommendation information set into at least two groups, wherein attributes of feature elements of the category of recommendation information in each of the at least two groups are mutually exclusive with attributes of feature elements of the category of recommendation information in other groups of the at least two groups, acquiring at least one type of result data corresponding to the recommendation information in each of the at least two groups, calculating a degree of difference between the result data of any two groups of the at least two groups for each type of result data, and determining that the feature elements of the category have significance to the result data if the degree of difference is greater than a first predetermined threshold.

According to an example of an embodiment of the present disclosure, the recommendation information processing unit is further configured to: in a case where it is determined that the feature elements of the category have significance to the result data, outputting information on the feature elements of the category, wherein the information includes evaluation information indicating an influence of the feature elements of the category on the result data, the recommendation information processing unit is further configured to generate a recommendation information analysis model for providing recommendation information production guidance based on the evaluation information.

According to an example of the embodiment of the present disclosure, wherein the information further includes viewer characteristic information indicating a common characteristic of viewers of recommendation information in a group of which result data is better among the two groups, the recommendation information processing unit is further configured to: and generating recommendation information oriented crowd recommendations based on the viewer characteristic information.

According to an example of an embodiment of the present disclosure, the recommendation information processing unit is further configured to: predicting a delivery effect of the new recommendation information by analyzing a plurality of categories of feature elements in the new recommendation information extracted by the feature extraction unit based on the output information on the different categories of feature elements.

According to an example of an embodiment of the present disclosure, wherein the at least two packets include a first packet and a second packet, wherein the recommendation information processing unit is further configured to: for each recommendation information of the plurality of recommendation information, dividing the recommendation information into a first group if a value or a number of feature elements of the category of the recommendation information satisfies a predetermined condition; otherwise, the recommendation information is divided into a second group.

According to an example of an embodiment of the present disclosure, wherein the feature extraction unit is further configured to: extracting feature elements of a plurality of content attribute categories in the recommendation information set by utilizing a deep learning model for the feature elements of the plurality of content attribute categories; and for each piece of recommendation information, generating that the recommendation information contains the confidence level of the extracted feature element of each content attribute category and the confidence level of the appearance position of the feature element of the content attribute category in the recommendation information.

According to an example of an embodiment of the present disclosure, the recommendation information processing unit is further configured to: adjusting the output information about the feature elements of different content attribute categories by adjusting the confidence level.

According to an example of an embodiment of the present disclosure, the feature extraction unit is further configured to: extracting a timestamp of each feature element in the feature element set; and the recommendation information processing unit is further configured to generate a recommendation information analysis model for providing recommendation information production guidance with time division based on the time stamp attached to each feature element of the recommendation information at the time of generating each recommendation information in the recommendation information set and the output information on the feature elements of different categories.

According to another aspect of the embodiments of the present disclosure, there is provided a recommendation information generating apparatus including: the feature extraction unit is configured to extract feature elements of a plurality of categories in the recommendation information to be optimized; a recommendation information generation unit configured to, for each of the plurality of categories of feature elements: dividing the recommendation information to be optimized into at least two groups, wherein the attribute of the characteristic element of the category of the recommendation information in each group of the at least two groups is mutually exclusive with the attribute of the characteristic element of the category of the recommendation information in other groups of the at least two groups, predicting at least one result data of the recommendation information to be optimized according to the preset evaluation information of the group aiming at the characteristic element of the category based on the group in which the recommendation information to be optimized is located, and optimizing the characteristic element of the category of the recommendation information to be optimized by using the predicted at least one result data.

According to another aspect of the embodiments of the present disclosure, there is provided a recommendation information processing apparatus including: one or more processors; and one or more memories having computer readable code stored therein, which when executed by the one or more processors, causes the one or more processors to perform the methods of the various aspects described above.

According to another aspect of embodiments of the present disclosure, there is provided a computer-readable storage medium having computer-readable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method described in the above aspects.

According to another aspect of embodiments of the present disclosure, there is provided a computer program product comprising computer readable instructions which, when executed by a processor, cause the processor to perform the method described in the above aspects.

By utilizing the recommendation information processing method, the recommendation information generating device, the recommendation information generating equipment, the computer readable storage medium and the computer program product according to the aspects, the influence of different types of feature elements on the recommendation information delivery result can be accurately analyzed and obtained; the system can provide recommendation information production guidance and recommendation information delivery oriented crowd recommendation; the putting effect of the recommendation information can be predicted; and optimizing and adjusting the recommendation information based on the predicted release effect to obtain a better recommendation information release effect.

Drawings

The above and other objects, features and advantages of the embodiments of the present disclosure will become more apparent by describing in more detail the embodiments of the present disclosure with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the disclosure and are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description serve to explain the principles of the disclosure and not to limit the disclosure. In the drawings, like reference numbers generally represent like parts or steps.

FIG. 1 shows a flow diagram of a recommendation information processing method according to an embodiment of the present disclosure;

FIG. 2 shows a schematic diagram of extracting feature elements according to an example of an embodiment of the present disclosure;

FIG. 3 shows a schematic diagram of grouping advertisements in a collection of advertisements, according to an example of an embodiment of the present disclosure;

FIG. 4 shows a schematic diagram of result data associating various packets according to an example of an embodiment of the present disclosure;

FIG. 5 shows a schematic diagram of calculating the variance of deviation of result data within various packets according to an example of an embodiment of the present disclosure;

FIG. 6 shows a flow diagram of a recommendation information generation method according to an embodiment of the present disclosure;

fig. 7 shows a schematic configuration diagram of a recommended information processing apparatus according to an embodiment of the present disclosure;

fig. 8 shows a schematic configuration diagram of a recommendation information generation apparatus according to an embodiment of the present disclosure;

fig. 9 shows a schematic diagram of an architecture of an exemplary computing device, according to an embodiment of the present disclosure.

Detailed Description

The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings in the embodiments of the present disclosure. It is to be understood that the described embodiments are merely exemplary of some, and not all, of the present disclosure. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without any inventive step, are intended to be within the scope of the present disclosure.

In the embodiment of the present disclosure, the recommendation information may be information that is pushed to the terminal device for display in a manner of a picture, a text, a video, or any combination thereof, for example. For example, the recommendation information may be any content such as advertisement, news push, etc., and the embodiment of the disclosure does not specifically limit this. In the following embodiments of the present disclosure, description will be made with an advertisement as an example of recommendation information.

Nowadays, the video advertisement is increasingly high in occupancy among advertisements. A video advertisement, for example, refers to an advertisement presented in the form of a video, which may be part or all of a video, e.g., a video advertisement may be attached to the beginning, end, or any other location of a particular video, or occupy the entire content of a video. Video advertisements have a number of features that are clearly distinguished from teletext advertisements due to their rich content attributes, complex production patterns, diversified presentation forms, etc. For example, for a teletext advertisement, the teletext advertisement can only be analyzed by a corresponding advertisement click rate, and it is difficult to know that the viewing duration, the viewing times, and the like of a viewer of the teletext advertisement can reflect more characteristics of the quality of the teletext advertisement; the video advertisements are different, and the watching time length, the watching times, the jumping-out time point, the conversion time point and the like of the video advertisements by the viewers can be collected to be used for analyzing the video advertisements, so that more and more complex advertisement putting result data are given to the video advertisements.

However, most of the conventional advertisement analysis methods are only suitable for the teletext advertisements, and the video advertisements cannot be accurately analyzed and evaluated. The present disclosure provides a recommendation information processing method that can effectively analyze recommendation information such as video advertisements, and can provide a recommendation information analysis model for recommendation information production guidance and targeted crowd recommendations for recommendation information delivery. In the following embodiments of the present disclosure, a video advertisement is described as an example, but it should be understood that some or all steps of the recommendation information processing method according to the embodiments of the present disclosure may be applied to any other recommendation information.

A recommendation information processing method according to an embodiment of the present disclosure is described below with reference to fig. 1. Fig. 1 shows a flow diagram of a recommendation information processing method 100 according to an embodiment of the present disclosure.

As shown in fig. 1, in step S110, a recommendation information set including a plurality of recommendation information is acquired. By way of example and not limitation, each recommendation information in the set of recommendation information may be an advertisement, such as a teletext advertisement or a video advertisement presented in the form of a video, and so forth. The recommendation information set may be obtained from an open-source database, for example, the recommendation information processing method according to the embodiment of the present disclosure may perform analysis by using existing data, for example, obtaining valuable advertisement information from historical advertisement data to perform advertisement analysis; alternatively, the recommendation information set may also be recommendation information to be delivered, for example, part or all of advertisements of a specific advertiser to be delivered, that is, the recommendation information processing method according to the embodiment of the disclosure may perform tracking analysis on the recommendation information to be delivered, so as to analyze and adjust the delivery effect of the recommendation information in real time.

In step S120, feature elements of a plurality of categories of each recommendation information in the recommendation information set are extracted to obtain a feature element set. Taking advertisements as an example, each advertisement is composed of a plurality of different categories of feature elements. For example, the teletext advertisement may comprise characteristic elements of the categories of color, shape, font, pattern, etc. The video advertisement may include richer feature elements, for example, the feature elements of the video advertisement may be divided into a basic attribute category and a content attribute category, where the basic attribute category may include various categories of feature elements such as frame rate, bitrate, size, resolution, saturation, definition, number of lenses, and the like of the advertisement, for example; the content attribute categories may include, for example, feature elements of various categories of shots, scenes, people, objects, presentations, etc. of advertisements. Here, for the video advertisement, the presentation form of the advertisement refers to a manner in which content included in the advertisement is presented, and for example, the presentation form of the video advertisement may include at least one of a dialog, a drill, a screen recording, a presentation (PPT), and the like, or any combination thereof.

The following describes the steps of extracting feature elements with reference to fig. 2 by taking a video advertisement as an example. Fig. 2 shows a schematic diagram of extracting feature elements according to an example of an embodiment of the present disclosure. As shown in fig. 2, for a video advertisement, extracting feature elements of multiple categories of each recommendation information in the set of recommendation information may include extracting feature elements of multiple base attribute categories and feature elements of multiple content attribute categories of each recommendation information.

Specifically, for example, for feature elements of a plurality of basic attribute categories of a video advertisement, feature elements of a frame rate, a bitrate, a size, a resolution, a saturation, a sharpness, and the like of the video advertisement may be extracted using a multimedia video processing tool such as FFMPEG, and a shot boundary detection algorithm such as a pixel comparison method, a dual threshold detection method, and the like may be used to extract the number of shots of the video advertisement. For feature elements of multiple content attribute categories of a video advertisement, among them, feature elements such as shots, scenes, people of the video advertisement may also be extracted using a shot boundary detection algorithm such as a pixel comparison method, a dual threshold detection method, and the like, but the present disclosure is not limited thereto. According to an example of an embodiment of the present disclosure, a deep learning model may be utilized to extract feature elements of a plurality of content attribute categories for advertisements in an advertisement collection. For example, taking the presentation form of the video advertisement as an example, different frames in the video advertisement may be first compared to extract key frames in the video advertisement, and then the key frames of the video advertisement may be processed using the deep learning model to identify the presentation form included in the video advertisement. Here, the deep learning model in the embodiments of the present disclosure may be a trained and fixed model for video analysis; alternatively, a deep learning model, such as a Convolutional Neural Network (CNN), a long-term short memory model (LSTM), etc., may also be built by using a neural network, and trained by using video samples from an open source video database, which is not specifically limited by the embodiment of the present disclosure.

Further, when extracting the feature elements of the plurality of content attribute categories of each recommendation information in the recommendation information set using the deep learning model, it is also possible to generate a confidence that the recommendation information contains the extracted feature elements of each content attribute category and a confidence of positions where the extracted feature elements of the plurality of content attribute categories appear in the recommendation information. For example, for a presentation form of a video advertisement, after determining that a certain video advertisement includes a conversation and a screen recording by using the deep learning model, a frame position or a time axis position where the conversation and the screen recording appear in the video advertisement may also be obtained, and a confidence that the video advertisement includes the conversation and the screen recording and a confidence that the conversation and the screen recording are at the frame position or the time axis position may be generated, so as to be used for subsequently adjusting and optimizing an advertisement analysis result, as will be described further below.

The feature elements of the plurality of categories of each piece of recommendation information in the extracted recommendation information set constitute a feature element set, and in the case of an advertisement, the feature element set may include values of each feature element of the resolution, saturation, sharpness, shot number, and the like of each advertisement in the plurality of advertisements. After the feature element set is extracted from the recommendation information set, for the feature element of each category in the feature element set, the influence of the feature element of the category on the result data (which may be referred to as result data) after the recommendation information is delivered may be determined through step S130. Step S130 may in turn comprise steps S131, S132 and S133.

In step S131, for each category of feature elements in the feature element set, the plurality of pieces of recommendation information in the recommendation information set are divided into at least two groups, where the attribute of the feature element of the category of recommendation information in each of the at least two groups is mutually exclusive from the attribute of the feature element of the category of recommendation information in the other of the at least two groups. Here, the attribute mutual exclusion of two feature elements may mean that values or numbers of the two feature elements have completely opposite attributes or attributes that are not compatible with each other. For example, for video advertisements, high and low resolution, high and low shot count, and low and recorded and non-recorded presentations, etc., are diametrically opposite attributes, and the advertisements in the set of advertisements may be divided into two groups accordingly, such that the advertisements in the two groups have diametrically opposite attributes. For another example, for video advertisements, where the resolution is high, the resolution is medium, and the resolution is low, the number of shots is large, the number of shots is medium, the number of shots is small, the number of screen recordings is large, the number of screen recordings is medium, the number of screen recordings is small, and the like are incompatible with each other, a plurality of advertisements in an advertisement collection may be divided into three groups accordingly such that the advertisements in the three groups have attributes that are incompatible with each other. It should be understood that, only by taking the advertisement as an example, an example of dividing the advertisement in the advertisement set into two groups and three groups is schematically listed here, but the embodiment of the present disclosure is not limited thereto, and the recommendation information in the recommendation information set may be divided into more groups according to actual needs.

Referring to fig. 3, the recommendation information is taken as an advertisement, and the advertisements in the advertisement set are divided into two groups. FIG. 3 shows a schematic diagram of grouping advertisements in a collection of advertisements, according to an example of an embodiment of the present disclosure. According to an example of an embodiment of the present disclosure, for a feature element of each category in a set of feature elements, for each advertisement in a set of advertisements, in case a value or a number of feature elements of the category of the advertisement satisfies a predetermined condition, the advertisement is divided into a first group; otherwise, the advertisement is divided into a second grouping. Here, the predetermined condition may be determined according to a value or a quantity distribution of the feature elements or an actual requirement, for example, the predetermined condition may be that the value or the quantity of the feature elements of the category of the advertisement is greater than, less than, or equal to a second predetermined threshold, or the predetermined condition may also be that the value or the quantity of the feature elements of the category of the advertisement belongs to a specific category, which is not specifically limited in the embodiment of the present disclosure.

For example, as shown in FIG. 3, for feature element category A, the advertisements in the set of advertisements may be divided into a first grouping A1 and a second grouping A2; for feature element category B, the advertisements in the set of advertisements may be divided into a first grouping B1 and a second grouping B2, and so on. For example, the feature element category a may be a resolution of a video advertisement, and when the resolution of a video advertisement in the advertisement set is greater than a second predetermined threshold, the video advertisement may be divided into a first group a 1; otherwise, when the resolution of the video advertisement is less than or equal to a second predetermined threshold, the video advertisement is divided into a second grouping A2. For another example, the feature element category B may be people in a video advertisement, and when the number of people in a certain video advertisement in the advertisement set is greater than a second predetermined threshold, the video advertisement may be divided into a first group B1; otherwise, when the number of people in the video advertisement is less than or equal to a second predetermined threshold, the video advertisement is divided into a second grouping B2.

In step S132, at least one type of result data corresponding to the recommendation information in each of the at least two groups is obtained. The at least one type of result data corresponding to the recommendation information in each group refers to various types of result data generated after the recommendation information of the group is released, and may include at least a part of release effect data, viewer behavior data, industry distribution data, and the like. For example, the advertisement effectiveness data may include an advertisement click rate, an advertisement conversion rate, and the like, wherein the advertisement click rate may refer to a ratio of a number of people clicking on a product link included in the advertisement to a total number of people watching the advertisement, and the advertisement conversion rate may refer to a ratio of a number of people acquiring a product by clicking on a product link included in the advertisement to a total number of people clicking on a product link included in the advertisement. Viewer behavior data may include the advertisement viewer's viewing duration, number of views, number of repeated views per day, play out rate, hop rate, conversion time point, conversion power, and so forth. Wherein, the broadcasting completion rate can be the ratio of the number of people who watch the video advertisement completely to the total number of people who open the video advertisement; the hop rate refers to the ratio of the number of people who jump out of the advertisement page due to reasons such as clicking on a product link contained in the advertisement, closing the advertisement page, etc., to the total number of people who view the advertisement; the conversion time point is a time point when a viewer clicks a product link contained in the advertisement in the process of watching the video advertisement; the conversion success rate is a ratio of the number of people who purchase the product by clicking a product link included in the advertisement to the total number of people who view the video advertisement. The industry distribution data represents the distribution characteristics of the industries to which the advertisement viewers belong, for example, the industry distribution data may include the proportion of the number of people in the advertisement viewers, the proportion of the number of people in the teacher industry, and the like, and further, in combination with the above-mentioned impression effect data and viewer behavior data, impression effect data and viewer behavior data in different industries may be obtained through statistics respectively.

According to an example of the embodiment of the present disclosure, after obtaining various result data of all recommendation information in the recommendation information set, the result data may be associated with respective groups obtained for different categories of feature elements, i.e., the respective groups for the different categories of feature elements may be associated with their corresponding result data. Referring to fig. 4, the advertisement set is divided into two groups for explanation. Fig. 4 shows a schematic diagram of result data associating individual packets according to an example of an embodiment of the present disclosure. As shown in fig. 4, for the first and second groups a1 and a2 of the advertisement sets divided for the feature element category a, the first and second groups a1 and a2 are respectively associated with a plurality of result data corresponding thereto including at least a part of the impression data, the viewer behavior data, and the industry distribution data; similarly, for the first group B1 and the second group B2 of the advertisement set divided for the feature element category B, the first group B1 and the second group B2 are respectively associated with a plurality of result data corresponding thereto, including at least a part of the placement effect data, the viewer behavior data, the industry distribution data, and the like, and so on, and will not be described herein again.

Thereafter, in step S133, for at least one type of result data corresponding to the recommendation information in each of at least two groups of the acquired recommendation information set, for each type of result data, a degree of difference between the result data of any two groups is calculated, and in a case where the degree of difference is greater than a first predetermined threshold, it is determined that the feature element of the current category has significance to the result data.

Specifically, for each type of result data, a total average value of the result data corresponding to any two groups is first calculated, for example, the result data corresponding to any two groups may be cumulatively summed, and the resulting cumulative sum may be divided by the total number of recommendation information in the two groups to obtain the total average value; then, in each of the two groups, a difference (which may be referred to as a back dispersion) between the result data of each recommendation information in the group and the total average value is calculated, and a variance (which may be referred to as a deviation variance) of the back dispersion of each recommendation information is calculated, for example, the deviation variances of the two groups may be referred to as a first deviation variance and a second deviation variance, respectively; a degree of difference of the first deviation variance and the second deviation variance is calculated, for example, the degree of difference may be an absolute difference of the first deviation variance and the second deviation variance. It should be noted that, although it is described above that the degree of difference between the result data of any two packets is calculated by the variance of deviation of the result data of each packet, the embodiments of the present disclosure are not limited thereto. The degree of difference between the result data of any two groups can also be calculated by an average value of the result data of each group or any other statistical index, which is not particularly limited by the embodiments of the present disclosure.

Referring to fig. 5, still taking the example of dividing the advertisement set into two groups, the specific steps of calculating the degree of difference will be further explained, and in this example, the group of advertisement sets divided for the feature element category a will be taken as an example.

Fig. 5 shows a schematic diagram of calculating the variance of deviation of result data within individual packets according to an example of an embodiment of the present disclosure. As shown in fig. 5, for feature element category a, the advertisements in the advertisement set are divided into a first group a1 and a second group a2, wherein the first group a1 and the second group a2 respectively have a corresponding plurality of result data including at least a portion of impression data, viewer behavior data, industry distribution data, and the like. In this step, for each kind of result data, the total average of the result data corresponding to the recommendation information of the first group a1 and the second group a2 is first calculated, and for example, the quotient of the total of the result data corresponding to the recommendation information of the groups a1, a2 and the total number of recommendation information of the groups a1, a2 may be taken as the total average; thereafter, a first variance of deviation of the result data for the first grouping A1 and a second variance of deviation of the result data for the second grouping A2 are calculated, respectively, e.g., a first variance of deviation A1-1 of advertisement click-through rates within the first grouping A1 and a second variance of deviation A2-1 of advertisement click-through rates within the second grouping A2 are calculated, respectively; a first deviation variance a1-2 of ad conversion rates within the first grouping a1 and a second deviation variance a2-2 of ad conversion rates within the second grouping a 2; a first variance of deviation a1-3 for viewer end broadcast rate within the first grouping a1 and a second variance of deviation a2-3 for viewer end broadcast rate within the second grouping a2, and so on.

In the case where there is a degree of difference between the result data of two groups of at least two groups greater than a first predetermined threshold, it is determined that the feature element of the current category has significance to the result data. In the embodiments of the present disclosure, significance means that the value of a feature element has a significant influence on the result data, i.e. the value of the result data can be significantly changed, for example as described above, such that the degree of difference between the result data for any two groups of the feature element is greater than a first predetermined threshold. Still referring to the example in FIG. 5 above, for example, if the degree of difference between the advertisement click rate within the first grouping A1 of the first deviation variance A1-1 and the second deviation variance A2-1 within the second grouping A2 is greater than a first predetermined threshold, the feature element category A may be determined to be significant for the advertisement conversion rate. As another example, if the degree of difference between the viewer end rate in the first deviation variance A1-3 within the first packet A1 and the second deviation variance A2-3 within the second packet A2 is greater than a first predetermined threshold, then the feature element category A may be determined to be significant to the viewer end rate. Here, the first predetermined threshold may be determined according to practical application requirements, and the embodiment of the present disclosure does not specifically limit this.

It should be noted that, although the advertisement set is described as being divided into two groups, the embodiments of the present disclosure are not limited thereto, and as described above, the advertisement set may be divided into more groups. For example, if the advertisement set is divided into a first group a1, a second group a2, and a second group A3 for the feature element category a, the degree of difference in the result data between any two of the three groups may be calculated separately and compared with a first predetermined threshold separately in step S133; alternatively, the two groups of the three groups with the best and worst variance of deviation of the result data may be selected, the degree of difference in variance of deviation between the two groups may be calculated and compared with a first predetermined threshold, and so on.

According to an example of the embodiment of the present disclosure, in the case where it is determined that a feature element of a certain category has significance to a certain result data, information about the feature element of the category, for example, evaluation information indicating an influence of the feature element of the category on the result data may be output. For example, as described above, in a case where it is determined that the feature element category a has significance to the advertisement click rate, evaluation information indicating the influence of the feature element category a on the advertisement click rate may be further output. Taking the feature element category a as the resolution of the video advertisements as an example, assuming that the resolution of the video advertisements in the first grouping a1 is high, the resolution of the video advertisements in the second grouping a1 is low, and the first deviation variance a1-1 of the advertisement click rate within the first grouping a1 is much smaller than the second deviation variance a2-1 of the advertisement click rate within the second grouping a2, i.e., the degree of difference between the two is greater than a first predetermined threshold, evaluation information such as "high resolution makes the deviation variance of the advertisement click rate smaller, performs better" or the like may be output. Alternatively, if the degree of difference is calculated using the average value instead of the variance in the above example, evaluation information such as "high resolution makes the average value of advertisement click rate larger" or the like may be output. According to an example of an embodiment of the present disclosure, the rating information may be used to produce a recommendation information analysis model, such as an advertisement analysis model, that provides recommendation information production guidance, as will be further described below.

Further, according to an example of an embodiment of the present disclosure, the output information about the feature elements of the category may further include viewer feature information indicating common features of viewers of the recommendation information in the group of better result data, for example, for generating recommendation information targeted crowd recommendations. The viewer characteristic information may be, for example, information indicating characteristics of the viewer's age, gender, industry, hobbies, region, etc., which may reflect the influence of the characteristic elements of the category on the viewing behavior of the population having the characteristics.

For example, in the above example where the feature element category a is the resolution of a video advertisement, assuming that the high resolution has been determined so that the deviation variance of the advertisement click rate is smaller, for the first and second groups a1 and a2 grouped for resolution, if the first deviation variance a1-1 of the advertisement click rate within the first group a1 is smaller, that is, the deviation of the advertisement click rate within the first group a1 from the total average value is smaller, the common features of viewers of the advertisements within the first group a1 may be extracted at this time. For example, all features of viewers of the advertisement within the first group a1 may be extracted and the probabilities of the respective features statistically analyzed, with one or more features having the highest probability being the common feature of those viewers. For example, assuming that the advertisement viewers in the first packet a1 are all extracted to have a common characteristic of "male, 18-25 years old, gaming industry", viewer characteristic information indicating the common characteristic may be output, which may indicate, for example, that a high resolution video advertisement has a large impact on the population having the characteristic of "male, 18-25 years old, gaming industry". Then when delivering video advertisements, if it is desired to obtain the best advertisement click-through rate, it may be recommended to select a crowd with the common features of "male, 18-25 years old, gaming industry" for targeted delivery to obtain better advertisement click-through revenue.

In the example of fig. 5, the analysis for the feature element category a is ended up to this point. Next, the analysis may be continued with the step S130 as described above for other feature element categories, for example, the analysis may be continued for the feature element category B to determine the influence of the feature element category B on different result data, and output corresponding evaluation information and viewer feature information; then, the feature element category C may be analyzed to determine the influence of the feature element category C on different result data, and output corresponding evaluation information and viewer feature information, and so on, which will not be described herein again.

According to an example of an embodiment of the present disclosure, evaluation information on different categories of feature elements output using a recommendation information processing method according to an embodiment of the present disclosure may be used to generate a recommendation information analysis model, such as an advertisement analysis model, that provides recommendation information production guidance. The recommendation information analysis model contains evaluation information indicating the influence of different classes of feature elements on different result data, such as the influence of resolution on advertisement click rate, the influence of advertisement presentation form on advertisement conversion rate, and the like. For example, a mapping table reflecting mapping relationships between the characteristic elements and the delivery result data representation may be generated based on the obtained various evaluation information, and the mapping table may be built in the advertisement analysis model. Therefore, when generating a new advertisement, the result data expected to be obtained, such as the click rate of the advertisement expected to be obtained, the conversion rate of the advertisement, etc., can be input into the advertisement analysis model, which guides the generation of the new advertisement by using the mapping table, for example, to output the new advertisement with the best expected performance. For example, the advertisement analysis model may guide the user to select which feature element categories and what values or quantities of the feature element categories are most likely to achieve the desired result data based on the mapping table, or may directly provide multiple advertisements with the best performance for the user to select, thereby facilitating the production of better quality advertisements. For another example, based on the obtained various evaluation information, an advertisement analysis model capable of intelligently generating new advertisements may be trained by using a neural network, for example, product attributes and advertisement requirements such as "high click-through rate" may be directly input, i.e., new advertisements with the best click-through rate may be generated, and so on. In addition, with the outputted information about different categories of feature elements, or an advertisement analysis model containing the information, for an unknown new advertisement, the impression effect of the new advertisement can be predicted and optimized by extracting and analyzing a plurality of categories of feature elements contained in the new advertisement, as will be described in further detail below.

According to an example of an embodiment of the present disclosure, viewer feature information regarding different categories of feature elements output using a recommendation information processing method according to an embodiment of the present disclosure may be used to generate recommendation information targeted crowd recommendations. After generating new recommendation information by using the recommendation information analysis model as described above, based on the categories of feature elements employed in the new recommendation information and the numerical values or numbers thereof, the viewer feature information of the feature elements related to the categories can be used to directionally recommend people most suitable for the new recommendation information, thereby contributing to obtaining more ideal benefits.

In addition, the recommended information processing method according to the embodiment of the present disclosure may further adjust the output information on the feature elements of different content attribute categories by adjusting the confidence degrees of the feature elements of the content attribute categories. For the characteristic elements of the content attribute category, such as the presentation form of the video advertisement, when the presentation form of the video advertisement is very complex, the presentation form of the video advertisement obtained through the deep learning model may have errors, thereby affecting the accuracy of the finally output information. At this time, as described above, when extracting the feature elements of a plurality of content attribute categories in the advertisement using the deep learning model, it is possible to generate a confidence that the advertisement contains the extracted feature elements of each content attribute category and a confidence that the feature elements of the content attribute category appear in the advertisement. For example, after information (e.g., rating information) about feature elements of a certain content attribute category is output, if the information is found to be less accurate after being verified in combination with actual conditions, the information may be adjusted and optimized by adjusting the confidence of the feature elements of the content attribute category to output the most accurate information about the feature elements of the content attribute category, thereby ensuring the accuracy of advertisement analysis.

In practical applications, the feature elements of the various categories used by the recommendation information may be updated frequently, or more specifically, different versions may exist for the feature elements of the same category. For example, in the recommendation information set acquired in step S110, different pieces of recommendation information may contain different versions of feature elements of the same category. For example, advertisement 1 and advertisement 2 in the advertisement set both contain characters, but advertisement 1 at an earlier time contains character a, and advertisement 2 at a later time contains a character updated to character b, and the influence of character a and character b on the result data may be different. If the advertisement analysis is performed only for the character feature category, and the update of the character version is not considered, the key information of the influence of the character update on the result data is missed. The traditional advertisement analysis method can only process static (namely, fixed and unchangeable) feature elements, and is difficult to provide a solution to the problems caused by updating the feature elements.

The recommendation information processing method 100 according to the embodiment of the present disclosure can effectively solve the above problems, and unlike the conventional advertisement analysis method that only depends on the service log after advertisement placement, the recommendation information processing method 100 according to the embodiment of the present disclosure can intervene in the placement process. Specifically, for example, in the advertisement generation phase, a timestamp may be attached to each category of feature elements included in the advertisement, which may indicate, for example, generation time, version, and the like of the category of feature elements for subsequent advertisement analysis. For example, in the case that the feature elements of different categories included in the advertisement set acquired in step S110 all have corresponding timestamps, when the feature elements of different categories of each advertisement in the advertisement set are extracted in step S120, the timestamp of the feature element of each category may also be extracted. At this time, for different versions indicated by the timestamps of the feature elements of each category, analysis may be performed in step S130 for the different versions respectively and corresponding information about the categories of the feature elements of the version may be output, and an advertisement analysis model with time divisions may be generated accordingly. That is, in the advertisement analysis model, for each category of feature elements, time division may be performed based on the timestamp thereof, and each version of the category of feature elements may be associated with the corresponding output information thereof, so that the influence of different versions of feature elements on the advertisement delivery result can be effectively obtained.

By using the recommendation information processing method according to the embodiment of the disclosure, the influence of different types of feature elements on the recommendation information delivery result can be accurately analyzed, recommendation information production guidance and targeted crowd recommendation of recommendation information delivery can be provided, and the delivery effect of the recommendation information can be predicted.

A recommendation information generation method according to an embodiment of the present disclosure is described below with reference to fig. 6. Fig. 6 shows a flowchart of a recommendation information generation method 600 according to an embodiment of the disclosure, where the recommendation information generation method 600 can optimize and adjust any recommendation information to be optimized. Since the details of the partial steps of the recommendation information generation method 600 are the same as those of the recommendation information processing method 100 described above with reference to fig. 1, the repeated description of the same contents is omitted here for the sake of simplicity.

As shown in fig. 6, in step S610, feature elements of a plurality of categories in recommendation information to be optimized are extracted. Here, the recommendation information to be optimized may be any information such as an advertisement, and may include, for example, a teletext advertisement, a video advertisement, and the like. The following description takes the recommendation information to be optimized as a video advertisement as an example.

According to an example of the embodiment of the present disclosure, extracting feature elements of multiple categories of recommendation information to be optimized may include extracting feature elements of multiple basic attribute categories of recommendation information to be optimized and feature elements of multiple content attribute categories. Specifically, taking the advertisement as an example, for the feature elements of the multiple basic attribute categories of the advertisement to be optimized, the multimedia video processing tool such as FFMPEG may be used to extract the feature elements of the advertisement to be optimized, such as frame rate, bitrate, size, resolution, saturation, sharpness, and the like, and the shot boundary detection algorithm such as pixel comparison method, dual threshold detection method, and the like may be used to extract the number of shots of the advertisement to be optimized. For feature elements of a plurality of content attribute categories of the advertisement to be optimized, wherein the feature elements such as a shot, a scene, a person, etc. of the advertisement to be optimized may also be extracted using a shot boundary detection algorithm such as a pixel comparison method, a dual threshold detection method, etc., but the present disclosure is not limited thereto.

According to an example of an embodiment of the present disclosure, a deep learning model may be utilized to extract feature elements of a plurality of content attribute categories of recommendation information to be optimized. For example, taking the presentation form of the advertisement to be optimized as an example, different frames in the advertisement to be optimized may be compared first to extract the key frames in the advertisement to be optimized, and then the key frames of the advertisement to be optimized may be processed by using the deep learning model to identify the presentation form included in the advertisement to be optimized. Here, the deep learning model in the embodiments of the present disclosure may be a trained and fixed model for video analysis; alternatively, a deep learning model, such as a Convolutional Neural Network (CNN), a long-term short memory model (LSTM), etc., may also be built by using a neural network, and trained by using video samples from an open source video database, which is not specifically limited by the embodiment of the present disclosure.

After extracting a plurality of categories of feature elements from the recommendation information to be optimized, for each category of feature elements, at least one result data of the recommendation information to be optimized may be predicted through step S620, and the feature elements of the category of the recommendation information to be optimized are optimized accordingly. Step S620 may in turn include steps S621, S622, and S623.

In step S621, for each category of feature elements, the recommendation information to be optimized is divided into a certain group of at least two groups, where the attribute of the category of feature element of the recommendation information in each group of the at least two groups is mutually exclusive from the attribute of the category of feature element of the recommendation information in other groups of the at least two groups. Here, the at least two groups may include two or more groups, which may be determined according to actual needs, and this is not specifically limited in this disclosure.

According to the example of the embodiment of the present disclosure, for each category of feature elements in a plurality of categories of feature elements, in the case that the value or the number of the feature elements of the category of the advertisement to be optimized satisfies a predetermined condition, the advertisement to be optimized is divided into a first group; otherwise, the advertisement to be optimized is divided into a second group. The predetermined condition may be determined according to a value or number distribution of the feature elements or an actual requirement, for example, the predetermined condition may be that the value or number of the feature elements of the category of the advertisement is greater than, less than, or equal to a second predetermined threshold, or the predetermined condition may also be that the value or number of the feature elements of the category of the advertisement belongs to a specific category, which is not specifically limited in the embodiment of the present disclosure. For example, for the resolution of the advertisement to be optimized, if the resolution of the advertisement to be optimized is greater than a second predetermined threshold, the advertisement to be optimized may be divided into a first grouping a 1; otherwise, when the resolution of the advertisement to be optimized is less than or equal to the second predetermined threshold, the advertisement to be optimized is classified into a second category a 2.

In step S622, based on the group in which the recommendation information to be optimized is located, at least one kind of result data of the recommendation information to be optimized is predicted from predetermined evaluation information on the group for the feature elements of the category. Here, the predetermined evaluation information may be, for example, information indicating an influence of different groups of feature elements for different categories on different result data, and as described above, the result data may include at least a part of the impression effect data, the viewer behavior data, and the industry distribution data. For example, in the above example, assuming that the advertisement to be optimized is divided into the first group a1 of high resolution for the resolution of the advertisement to be optimized, and the predetermined evaluation information for the first group a1 of resolution indicates the influence of the first group a1 of resolution on different result data, for example, the high resolution makes the deviation variance of the advertisement click rate small, the average value high, and the like, at least one of the result data of the advertisement to be optimized can be predicted accordingly. The predetermined evaluation information may be obtained by the recommendation information processing method 100 as described above, and may be, for example, evaluation information about feature elements of different categories included in a recommendation information analysis model generated by the recommendation information processing method 100, or may also be information generated in advance by another method, which is not specifically limited by the embodiment of the present disclosure.

After at least one result data of the recommendation information to be optimized for the feature element of the current category is predicted, in step S623, the feature element of the category of the recommendation information to be optimized may be optimized by using the predicted at least one result data. For example, for the resolution of the advertisement to be optimized, assuming that the advertisement to be optimized is divided into the first group a1 with low resolution, and the result data predicted according to step S622 is not good enough, for example, the deviation variance of the predicted advertisement click rate is large and the average value is small, then the resolution of the advertisement to be optimized can be optimized and adjusted at this time by increasing it.

By using the recommendation information generation method according to the embodiment, optimization adjustment can be performed on different types of feature elements contained in any recommendation information to be optimized, so that the recommendation information to be optimized can have a better release effect.

A recommended information processing apparatus according to an embodiment of the present disclosure is described below with reference to fig. 7. Fig. 7 shows a schematic configuration diagram of a recommended information processing apparatus 700 according to an embodiment of the present disclosure. As shown in fig. 7, the recommended information processing apparatus 700 includes a recommended information acquisition unit 710, a feature extraction unit 720, and a recommended information processing unit 730. The recommendation information processing 700 may include other components in addition to the three units, however, since these components are not related to the contents of the embodiments of the present disclosure, illustration and description thereof are omitted herein.

The recommendation information acquisition unit 710 is configured to acquire a recommendation information set containing a plurality of recommendation information. By way of example and not limitation, each recommendation information in the set of recommendation information may be an advertisement, such as a video advertisement, i.e., an advertisement presented in the form of a video. The recommendation information set may be obtained from an open source advertisement database, for example, the recommendation information processing apparatus according to the embodiment of the present disclosure may perform analysis using existing data, for example, obtaining valuable advertisement information from historical advertisement data to perform advertisement analysis; alternatively, the recommendation information set may also be recommendation information to be delivered, for example, part or all of advertisements of a specific advertiser to be delivered, that is, the recommendation information processing apparatus according to the embodiment of the disclosure may perform tracking analysis on the recommendation information to be delivered, so as to analyze and adjust the delivery effect of the recommendation information in real time.

The feature extraction unit 720 is configured to extract feature elements of a plurality of categories of each recommendation information in the recommendation information set to obtain a feature element set. Taking advertisements as an example, each advertisement is composed of a plurality of different categories of feature elements. For example, the teletext advertisement may comprise characteristic elements of the categories of color, shape, font, pattern, etc. The video advertisement may include richer feature elements, for example, the feature elements of the video advertisement may be divided into a basic attribute category and a content attribute category, where the basic attribute category may include various categories of feature elements such as frame rate, bitrate, size, resolution, saturation, definition, number of lenses, and the like of the advertisement, for example; the content attribute categories may include, for example, feature elements of various categories of shots, scenes, people, objects, presentations, etc. of advertisements. Here, for the video advertisement, the presentation form of the advertisement refers to a manner in which content included in the advertisement is presented, and for example, the presentation form of the video advertisement may include at least one of a dialog, a drill, a screen recording, a presentation (PPT), and the like, or any combination thereof.

The steps of extracting feature elements by the feature extraction unit 720 are described with reference to fig. 2 by taking a video advertisement as an example. As shown in fig. 2, for a video advertisement, extracting feature elements of multiple categories of each recommendation information in the set of recommendation information may include extracting feature elements of multiple base attribute categories and feature elements of multiple content attribute categories of each recommendation information.

Specifically, for example, for feature elements of a plurality of basic attribute categories of a video advertisement, the feature extraction unit 720 may extract the feature elements of the video advertisement, such as frame rate, bitrate, size, resolution, saturation, sharpness, and the like, using a multimedia video processing tool, such as FFMPEG, and may extract the number of lenses of the video advertisement using a lens boundary detection algorithm, such as a pixel comparison method, a dual threshold detection method, and the like. For feature elements of multiple content attribute categories of a video advertisement, among them, feature elements such as shots, scenes, people of the video advertisement may also be extracted using a shot boundary detection algorithm such as a pixel comparison method, a dual threshold detection method, and the like, but the present disclosure is not limited thereto. According to an example of an embodiment of the present disclosure, the feature extraction unit 720 may utilize a deep learning model to extract feature elements of a plurality of content attribute categories of advertisements in an advertisement set. For example, taking the presentation form of the video advertisement as an example, the feature extraction unit 720 may first compare different frames in the video advertisement to extract key frames in the video advertisement, and then process the key frames of the video advertisement using the deep learning model to identify the presentation form included in the video advertisement. Here, the deep learning model in the embodiments of the present disclosure may be a trained and fixed model for video analysis; alternatively, a deep learning model, such as a Convolutional Neural Network (CNN), a long-term short memory model (LSTM), etc., may also be built by using a neural network, and trained by using video samples from an open source video database, which is not specifically limited by the embodiment of the present disclosure.

Furthermore, the feature extraction unit 720 may be further configured to, when extracting the feature elements of the plurality of content attribute categories of each recommendation information in the recommendation information set using the deep learning model, generate that the recommendation information contains the confidence of the extracted feature elements of each content attribute category and the confidence of the positions where the extracted feature elements of the plurality of content attribute categories appear in the recommendation information. For example, for a presentation form of a video advertisement, the feature extraction unit 720, after determining that a certain video advertisement includes a conversation and a screen recording by using the deep learning model, may further obtain a frame position or a time axis position where the conversation and the screen recording appear in the video advertisement, and generate a confidence that the video advertisement includes the conversation and the screen recording, and a confidence that the conversation and the screen recording are at the frame position or the time axis position, for subsequent adjustment and optimization of an advertisement analysis result, as will be further described below.

The feature elements of the plurality of categories of each piece of recommendation information in the extracted recommendation information set constitute a feature element set, and in the case of an advertisement, the feature element set may include values of each feature element of the resolution, saturation, sharpness, shot number, and the like of each advertisement in the plurality of advertisements. After extracting the feature element set from the recommendation information set, for the feature element of each category in the feature element set, the recommendation information processing unit 730 may determine an influence of the feature element of the category on result data (which may be referred to as result data) after the recommendation information is delivered, which is described below.

First, the recommendation information processing unit 730 is configured to divide the plurality of pieces of recommendation information in the recommendation information set into at least two groups for each category of feature elements in the feature element set, wherein an attribute of the feature element of the category of recommendation information in each of the at least two groups is mutually exclusive from an attribute of the feature element of the category of recommendation information in the other of the at least two groups. Here, the attribute mutual exclusion of two feature elements may mean that values or numbers of the two feature elements have completely opposite attributes or attributes that are not compatible with each other. For example, for video advertisements, high and low resolution, high and low shot count, and low and recorded and non-recorded presentations, etc., are diametrically opposite attributes, and the advertisements in the set of advertisements may be divided into two groups accordingly, such that the advertisements in the two groups have diametrically opposite attributes. For another example, for video advertisements, where the resolution is high, the resolution is medium, and the resolution is low, the number of shots is large, the number of shots is medium, the number of shots is small, the number of screen recordings is large, the number of screen recordings is medium, the number of screen recordings is small, and the like are incompatible with each other, a plurality of advertisements in an advertisement collection may be divided into three groups accordingly such that the advertisements in the three groups have attributes that are incompatible with each other. It should be understood that, only by way of example, the advertisements in the advertisement set are schematically listed as examples of dividing the advertisements in the advertisement set into two groups and three groups, but the embodiments of the present disclosure are not limited thereto, and the advertisements in the advertisement set may be divided into more groups according to actual needs.

Thereafter, the recommendation information processing unit 730 is configured to acquire at least one type of result data corresponding to the recommendation information in each of the at least two groups. The at least one type of result data corresponding to the recommendation information in each group refers to various types of result data generated after the recommendation information of the group is released, and may include at least a part of release effect data, viewer behavior data, industry distribution data, and the like. For example, the advertisement effectiveness data may include an advertisement click rate, an advertisement conversion rate, and the like, wherein the advertisement click rate may refer to a ratio of a number of people clicking on a product link included in the advertisement to a total number of people watching the advertisement, and the advertisement conversion rate may refer to a ratio of a number of people acquiring a product by clicking on a product link included in the advertisement to a total number of people clicking on a product link included in the advertisement. Viewer behavior data may include the advertisement viewer's viewing duration, number of views, number of repeated views per day, play out rate, hop rate, conversion time point, conversion power, and so forth. Wherein, the broadcasting completion rate can be the ratio of the number of people who watch the video advertisement completely to the total number of people who open the video advertisement; the hop rate refers to the ratio of the number of people who jump out of the advertisement page due to reasons such as clicking on a product link contained in the advertisement, closing the advertisement page, etc., to the total number of people who view the advertisement; the conversion time point is a time point when a viewer clicks a product link contained in the advertisement in the process of watching the video advertisement; the conversion success rate is a ratio of the number of people who purchase the product by clicking a product link included in the advertisement to the total number of people who view the video advertisement. The industry distribution data represents the distribution characteristics of the industries to which the advertisement viewers belong, for example, the industry distribution data may include the proportion of the number of people in the advertisement viewers, the proportion of the number of people in the teacher industry, and the like, and further, in combination with the above-mentioned impression effect data and viewer behavior data, impression effect data and viewer behavior data in different industries may be obtained through statistics respectively.

According to an example of the embodiment of the present disclosure, after the recommendation information processing unit 730 acquires various result data of all recommendation information in the recommendation information set, the result data may be associated with respective groups obtained for different categories of feature elements, that is, the respective groups for the different categories of feature elements may be associated with their corresponding result data. Referring to fig. 4, the advertisement set is divided into two groups for explanation. Fig. 4 shows a schematic diagram of result data associating individual packets according to an example of an embodiment of the present disclosure. As shown in fig. 4, for the first and second groups a1 and a2 of the advertisement sets divided for the feature element category a, the first and second groups a1 and a2 are respectively associated with a plurality of result data corresponding thereto including at least a part of the impression data, the viewer behavior data, and the industry distribution data; similarly, for the first group B1 and the second group B2 of the advertisement set divided for the feature element category B, the first group B1 and the second group B2 are respectively associated with a plurality of result data corresponding thereto, including at least a part of the placement effect data, the viewer behavior data, the industry distribution data, and the like, and so on, and will not be described herein again.

Thereafter, the recommendation information processing unit 730 is configured to calculate, for each of at least one type of result data corresponding to each of at least two groups of the acquired recommendation information set, a degree of difference between the result data of any two groups, and determine that the feature element of the current category has significance to the result data in a case where the degree of difference is greater than a first predetermined threshold.

Specifically, for each type of result data, the recommendation information processing unit 730 may first calculate a total average value of the result data corresponding to any two groups, for example, may cumulatively sum the result data corresponding to any two groups, and divide the resulting cumulative sum by the total number of recommendation information in the two groups to obtain the total average value; then, in each of the two groups, a difference (which may be referred to as a back dispersion) between the result data of each recommendation information in the group and the total average value is calculated, and a variance (which may be referred to as a deviation variance) of the back dispersion of each recommendation information is calculated, for example, the deviation variances of the two groups may be referred to as a first deviation variance and a second deviation variance, respectively; a degree of difference of the first deviation variance and the second deviation variance is calculated, for example, the degree of difference may be an absolute difference of the first deviation variance and the second deviation variance. It should be noted that, although it is described above that the degree of difference between the result data of any two packets is calculated by the variance of deviation of the result data of each packet, the embodiments of the present disclosure are not limited thereto. The degree of difference between the result data of any two groups can also be calculated by an average value of the result data of each group or any other statistical index, which is not particularly limited by the embodiments of the present disclosure.

Referring to fig. 5, still taking the example of dividing the advertisement set into two groups, the specific step of the recommendation information processing unit 730 calculating the degree of difference will be further explained, and in this example, the group of the advertisement set divided for the feature element category a is taken as an example.

As shown in fig. 5, for feature element category a, the advertisements in the advertisement set are divided into a first group a1 and a second group a2, wherein the first group a1 and the second group a2 respectively have a corresponding plurality of result data including at least a portion of impression data, viewer behavior data, industry distribution data, and the like. For each kind of result data, the recommendation information processing unit 730 first calculates the total average of the result data of the recommendation information of the first and second groups a1, a2, for example, the quotient of the total of the result data of the recommendation information of the groups a1, a2 and the total number of recommendation information of the groups a1, a2 may be taken as the above total average; thereafter, a first variance of deviation of the result data for the first grouping A1 and a second variance of deviation of the result data for the second grouping A2 are calculated, respectively, e.g., a first variance of deviation A1-1 of advertisement click-through rates within the first grouping A1 and a second variance of deviation A2-1 of advertisement click-through rates within the second grouping A2 are calculated, respectively; a first deviation variance a1-2 of ad conversion rates within the first grouping a1 and a second deviation variance a2-2 of ad conversion rates within the second grouping a 2; a first variance of deviation a1-3 for viewer end broadcast rate within the first grouping a1 and a second variance of deviation a2-3 for viewer end broadcast rate within the second grouping a2, and so on.

In a case where there is a degree of difference of the result data corresponding to two groups in at least two groups, which is greater than a first predetermined threshold, the recommendation information processing unit 730 may determine that the feature element of the current category has significance to the result data. In the embodiments of the present disclosure, significance means that the value of a feature element has a significant influence on the result data, i.e. the value of the result data can be significantly changed, for example as described above, such that the degree of difference between the result data for any two groups of the feature element is greater than a first predetermined threshold. Still referring to the example in FIG. 5 above, for example, if the degree of difference between the advertisement click rate within the first grouping A1 of the first deviation variance A1-1 and the second deviation variance A2-1 within the second grouping A2 is greater than a first predetermined threshold, the feature element category A may be determined to be significant for the advertisement conversion rate. As another example, if the viewer end rate differs by more than a first predetermined threshold between a first deviation variance a1-3 of the first packet a1 within the first packet a1 and a second deviation variance a2-3 within the second packet a2, the feature element category a may be determined to be significant to the viewer end rate. Here, the first predetermined threshold may be determined according to practical application requirements, and the embodiment of the present disclosure does not specifically limit this.

According to an example of an embodiment of the present disclosure, in a case where it is determined that a feature element of a certain category has significance to a certain result data, the recommendation information processing unit 730 may output information about the feature element of the category, for example, evaluation information indicating an influence of the feature element of the category on the result data. For example, as described above, in a case where it is determined that the feature element category a has significance to the advertisement click rate, evaluation information indicating the influence of the feature element category a on the advertisement click rate may be further output. Taking the feature element category a as the resolution of the video advertisements as an example, assuming that the resolution of the video advertisements in the first grouping a1 is high, the resolution of the video advertisements in the second grouping a1 is low, and the first deviation variance a1-1 of the advertisement click rate within the first grouping a1 is much smaller than the second deviation variance a2-1 of the advertisement click rate within the second grouping a2, i.e., the degree of difference between the two is greater than a first predetermined threshold, evaluation information such as "high resolution makes the deviation variance of the advertisement click rate smaller, performs better" or the like may be output. Alternatively, if the degree of difference is calculated using the average value instead of the variance in the above example, evaluation information such as "high resolution makes the average value of advertisement click rate larger" or the like may be output. According to an example of an embodiment of the present disclosure, the rating information may be used to produce a recommendation information analysis model that provides recommendation information production guidance, e.g., an advertisement analysis model as will be further described below.

Further, according to an example of an embodiment of the present disclosure, the information about the feature elements of the category that may be output by the recommendation information processing unit 730 may further include viewer feature information indicating common features of viewers of recommendation information in a group of better result data, for example, for generating recommendation information targeted crowd recommendations. The viewer characteristic information may be, for example, information indicating characteristics of the viewer's age, gender, industry, hobbies, region, etc., which may reflect the influence of the characteristic elements of the category on the viewing behavior of the population having the characteristics.

For example, in the above example where the feature element category a is the resolution of a video advertisement, assuming that the high resolution has been determined so that the deviation variance of the advertisement click rate is smaller, for the first and second groups a1 and a2 grouped for resolution, if the first deviation variance a1-1 of the advertisement click rate within the first group a1 is smaller, that is, the deviation of the advertisement click rate within the first group a1 from the total average value is smaller, the common features of viewers of the advertisements within the first group a1 may be extracted at this time. For example, all features of viewers of the advertisement within the first group a1 may be extracted, and the probabilities of the respective features are statistically analyzed, and one or more features with the highest probability are taken as common features for these viewers. Assuming that the advertisement viewers in the first packet a1 all have a common characteristic of "male, 18-25 years old, gaming industry", viewer characteristic information indicating the common characteristic may be output, which may indicate, for example, that a high resolution video advertisement has a large impact on the population having the characteristic of "male, 18-25 years old, gaming industry". According to examples of embodiments of the present disclosure, viewer characteristic information may be used to generate advertisement targeted crowd recommendations, as will be further described below. Then when delivering video advertisements, if it is desired to obtain the best advertisement click-through rate, it may be recommended to select a crowd with the common features of "male, 18-25 years old, gaming industry" for targeted delivery to obtain better advertisement click-through revenue.

In the example of fig. 5, up to this point, the recommendation information processing unit 730 may end the analysis for the feature element category a. Next, the recommendation information processing unit 730 may continue the analysis with the step S130 as described above for other feature element categories, for example, may continue the analysis for the feature element category B to determine the influence of the feature element category B on different result data and output corresponding evaluation information and viewer feature information; then, the feature element category C may be analyzed to determine the influence of the feature element category C on different result data, and output corresponding evaluation information and viewer feature information, and so on, which will not be described herein again.

According to an example of an embodiment of the present disclosure, evaluation information on different categories of feature elements output with a recommendation information processing apparatus according to an embodiment of the present disclosure may be used to generate a recommendation information analysis model, such as an advertisement analysis model, that provides recommendation information production guidance. The recommendation information analysis model contains evaluation information indicating the influence of different classes of feature elements on different result data, such as the influence of resolution on advertisement click rate, the influence of advertisement presentation form on advertisement conversion rate, and the like. For example, a mapping table reflecting mapping relationships between the characteristic elements and the delivery result data representation may be generated based on the obtained various evaluation information, and the mapping table may be built in the advertisement analysis model. Therefore, when generating a new advertisement, the result data expected to be obtained, such as the click rate of the advertisement expected to be obtained, the conversion rate of the advertisement, etc., can be input into the advertisement analysis model, which guides the generation of the new advertisement by using the mapping table, for example, to output the new advertisement with the best expected performance. For example, the advertisement analysis model may direct which feature element categories are selected based on the mapping table, and what values or quantities of the feature element categories are most likely to achieve the desired result data, or may directly provide multiple advertisements with the best performance for the user to select, thereby facilitating the production of better quality advertisements. For another example, based on the obtained various evaluation information, an advertisement analysis model capable of intelligently generating new advertisements may be trained by using a neural network, for example, product attributes and advertisement requirements such as "high click-through rate" may be directly input, i.e., new advertisements with the best click-through rate may be generated, and so on. In addition, with the output information about the different categories of feature elements, or an advertisement analysis model containing the information, for an unknown new advertisement, the impression effect of the new advertisement can be predicted by extracting and analyzing the categories of feature elements contained in the new advertisement, as will be described in further detail below.

According to an example of an embodiment of the present disclosure, viewer characteristic information regarding different categories of characteristic elements output by a recommendation information processing apparatus according to an embodiment of the present disclosure may be used to generate recommendation information targeted crowd recommendations. After generating new recommendation information by using the recommendation information analysis model as described above, based on the categories of feature elements employed in the new recommendation information and the numerical values or numbers thereof, the viewer feature information of the feature elements related to the categories can be used to directionally recommend people most suitable for the new recommendation information, thereby contributing to obtaining more ideal benefits.

In addition, the recommended information processing apparatus according to the embodiment of the present disclosure may also adjust the output information on the feature elements of different content attribute categories by adjusting the confidence degrees of the feature elements of the content attribute categories. For the characteristic elements of the content attribute category, such as the presentation form of the video advertisement, when the presentation form of the video advertisement is very complex, the presentation form of the video advertisement obtained through the deep learning model may have errors, thereby affecting the accuracy of the finally output information. At this time, as described above, when extracting the feature elements of a plurality of content attribute categories in the advertisement using the deep learning model, it is possible to generate a confidence that the advertisement contains the extracted feature elements of each content attribute category and a confidence that the feature elements of the content attribute category appear in the advertisement. For example, after information (e.g., rating information) about feature elements of a certain content attribute category is output, if the information is found to be less accurate after being verified in combination with actual conditions, the information may be adjusted and optimized by adjusting the confidence of the feature elements of the content attribute category to output the most accurate information about the feature elements of the content attribute category, thereby ensuring the accuracy of advertisement analysis.

In practical applications, the feature elements of the various categories used by the recommendation information may be updated frequently, or more specifically, different versions may exist for the feature elements of the same category. For example, in the recommendation information set acquired by the recommendation information acquisition unit 710, different pieces of recommendation information may contain different versions of feature elements of the same category. For example, advertisement 1 and advertisement 2 in the advertisement set both contain characters, but advertisement 1 at an earlier time contains character a, and advertisement 2 at a later time contains a character updated to character b, and the influence of character a and character b on the result data may be different. If the advertisement analysis is performed only for the character feature category, and the update of the character version is not considered, the key information of the influence of the character update on the result data is missed. The traditional advertisement analysis method can only process static (namely, fixed and unchangeable) feature elements, and is difficult to provide a solution to the problems caused by updating the feature elements.

The recommendation information processing apparatus 700 according to the embodiment of the present disclosure can effectively solve the above-mentioned problems, and unlike the conventional advertisement analysis method that only relies on the service log after advertisement placement, the recommendation information processing apparatus 700 according to the embodiment of the present disclosure can intervene in the placement process. Specifically, for example, in the advertisement generation phase, a timestamp may be attached to each category of feature elements included in the advertisement, which may indicate, for example, generation time, version, and the like of the category of feature elements for subsequent advertisement analysis. For example, in the case where feature elements of different categories included in the advertisement set acquired by the recommendation information acquisition unit 710 all have corresponding timestamps, when the feature extraction unit 720 extracts feature elements of different categories of each advertisement in the advertisement set, the timestamp of the feature element of each category may also be extracted. At this time, for different versions indicated by the timestamps of the feature elements of each category, the recommendation information processing unit 730 may analyze and output corresponding information on the category of the feature elements of the version for the different versions, respectively, and generate an advertisement analysis model having time divisions according thereto. That is, in the advertisement analysis model, for each category of feature elements, time division may be performed based on the timestamp thereof, and each version of the category of feature elements may be associated with the corresponding output information thereof, so that the influence of different versions of feature elements on the advertisement delivery result can be effectively obtained.

By using the recommendation information processing device according to the embodiment of the disclosure, the influence of different types of feature elements on the recommendation information delivery result can be accurately analyzed, recommendation information production guidance and targeted crowd recommendation of recommendation information delivery can be provided, and the delivery effect of the recommendation information can be predicted.

A recommendation information generation apparatus according to an embodiment of the present disclosure is described below with reference to fig. 8. Fig. 8 shows a schematic structural diagram of a recommendation information generation apparatus 800 according to an embodiment of the present disclosure. As shown in fig. 8, recommendation information generation apparatus 700 includes feature extraction unit 810 and recommendation information generation unit 820. The recommendation information generation apparatus 800 may include other components in addition to the two units, however, since these components are not related to the contents of the embodiments of the present disclosure, illustration and description thereof are omitted herein.

As shown in fig. 8, the feature extraction unit 810 is configured to extract feature elements of a plurality of categories in the recommendation information to be optimized. Here, the recommendation information to be optimized may be any information such as an advertisement, and may include, for example, a teletext advertisement, a video advertisement, and the like. The following description takes the recommendation information to be optimized as a video advertisement as an example.

According to an example of the embodiment of the present disclosure, extracting feature elements of multiple categories of recommendation information to be optimized may include extracting feature elements of multiple basic attribute categories of recommendation information to be optimized and feature elements of multiple content attribute categories. Specifically, taking an advertisement as an example, for feature elements of a plurality of basic attribute categories of the advertisement to be optimized, the feature extraction unit 810 may extract the feature elements of the frame rate, bitrate, size, resolution, saturation, sharpness, and the like of the advertisement to be optimized using a multimedia video processing tool such as FFMPEG and the like, and may extract the number of lenses of the advertisement to be optimized using a lens boundary detection algorithm such as a pixel comparison method, a dual threshold detection method, and the like. For feature elements of a plurality of content attribute categories of the advertisement to be optimized, wherein the feature elements such as a shot, a scene, a person, etc. of the advertisement to be optimized may also be extracted using a shot boundary detection algorithm such as a pixel comparison method, a dual threshold detection method, etc., but the present disclosure is not limited thereto.

According to an example of an embodiment of the present disclosure, the feature extraction unit 810 may extract feature elements of a plurality of content attribute categories of recommendation information to be optimized using a deep learning model. For example, taking the presentation form of the advertisement to be optimized as an example, the feature extraction unit 810 may first compare different frames in the advertisement to be optimized to extract key frames in the advertisement to be optimized, and then process the key frames of the advertisement to be optimized by using a deep learning model to identify the presentation form included in the advertisement to be optimized. Here, the deep learning model in the embodiments of the present disclosure may be a trained and fixed model for video analysis; alternatively, a deep learning model, such as a Convolutional Neural Network (CNN), a long-term short memory model (LSTM), etc., may also be built by using a neural network, and trained by using video samples from an open source video database, which is not specifically limited by the embodiment of the present disclosure.

After extracting a plurality of categories of feature elements from the recommendation information to be optimized, for each category of feature elements, the recommendation information generation unit 820 is configured to predict at least one result data of the recommendation information to be optimized, and optimize the category of feature elements of the recommendation information to be optimized according to the result data, as described below.

First, the recommendation information generating unit 820 is configured to divide the recommendation information to be optimized into a certain group of at least two groups for each class of feature elements, wherein the attribute of the class of feature elements of the recommendation information in each of the at least two groups is mutually exclusive from the attribute of the class of feature elements of the recommendation information in the other of the at least two groups. Here, the at least two groups may include two or more groups, which may be determined according to actual needs, and this is not specifically limited in this disclosure.

Thereafter, the recommendation information generation unit 820 is configured to predict at least one result data of the recommendation information to be optimized from predetermined evaluation information on the group of the feature elements for the category based on the group in which the recommendation information to be optimized is located. Here, the predetermined evaluation information may be, for example, information indicating an influence of different groups of feature elements for different categories on different result data, and as described above, the result data may include at least a part of the impression effect data, the viewer behavior data, and the industry distribution data. For example, in the above example, assuming that the advertisement to be optimized is divided into the first group a1 of high resolution for the resolution of the advertisement to be optimized, and the predetermined evaluation information for the first group a1 of resolution indicates the influence of the first group a1 of resolution on different result data, for example, the high resolution makes the deviation variance of the advertisement click rate small, the average value high, and the like, at least one of the result data of the advertisement to be optimized can be predicted accordingly. The predetermined evaluation information may be obtained by the recommendation information processing method 100 as described above, and may be, for example, evaluation information about feature elements of different categories included in the advertisement analysis model generated by the recommendation information processing method 100, or may also be information generated in advance by other methods, which is not specifically limited by the embodiment of the present disclosure.

After predicting at least one result data of the advertisement to be optimized for the feature element of the current category, the recommendation information generation unit 820 is configured to optimize the feature element of the category of the recommendation information to be optimized by using the predicted at least one result data. For example, for the resolution of the advertisement to be optimized, assuming that the advertisement to be optimized is divided into the first packet a1 with low resolution, and the predicted result data is not good enough, for example, the deviation variance of the predicted advertisement click rate is large and the average value is small, then the resolution of the advertisement to be optimized can be optimized and adjusted by increasing it.

By using the recommendation information generation device according to the embodiment, optimization adjustment can be performed on different types of feature elements included in any recommendation information to be optimized, so that the recommendation information to be optimized can have a better advertisement putting effect.

Furthermore, devices (e.g., recommendation information processing devices, recommendation information generation devices, etc.) according to embodiments of the present disclosure may also be implemented by way of the architecture of an exemplary computing device shown in fig. 9. Fig. 9 shows a schematic diagram of an architecture of an exemplary computing device, according to an embodiment of the present disclosure. As shown in fig. 9, computing device 900 may include a bus 910, one or more CPUs 920, a Read Only Memory (ROM)930, a Random Access Memory (RAM)940, a communication port 950 connected to a network, input/output components 960, a hard disk 970, and the like. Storage devices in the computing device 900, such as the ROM 930 or the hard disk 970, may store various data or files used in computer processing and/or communications as well as program instructions executed by the CPU. Computing device 900 may also include a user interface 980. Of course, the architecture shown in FIG. 9 is merely exemplary, and one or more components of the computing device shown in FIG. 9 may be omitted when implementing different devices, as desired. The device according to the embodiments of the present disclosure may be configured to execute the recommendation information processing method or the recommendation information generation method according to the above-described respective embodiments of the present disclosure, or to implement the recommendation information processing apparatus or the recommendation information generation apparatus according to the above-described respective embodiments of the present disclosure.

Embodiments of the present disclosure may also be implemented as a computer-readable storage medium. A computer readable storage medium according to an embodiment of the present disclosure has computer readable instructions stored thereon. The recommended information processing method or the recommended information processing generation method according to the embodiment of the present disclosure described with reference to the above drawings may be performed when the computer readable instructions are executed by a processor. Computer-readable storage media include, but are not limited to, volatile memory and/or nonvolatile memory, for example. Volatile memory can include, for example, Random Access Memory (RAM), cache memory (or the like). The non-volatile memory may include, for example, Read Only Memory (ROM), a hard disk, flash memory, and the like.

There is also provided, in accordance with an embodiment of the present disclosure, a computer program product or computer program, including computer readable instructions, the computer readable instructions being stored in a computer readable storage medium. The processor of the computer device may read the computer readable instructions from the computer readable storage medium, and execute the computer readable instructions, so that the computer device executes the recommendation information processing method or the recommendation information generation method described in the above embodiments.

Those skilled in the art will appreciate that the disclosure of the present disclosure is susceptible to numerous variations and modifications. For example, the various devices or components described above may be implemented in hardware, or may be implemented in software, firmware, or a combination of some or all of the three.

Furthermore, as used in this disclosure and in the claims, the terms "a," "an," "the," and/or "the" are not intended to be inclusive in the singular, but rather are inclusive in the plural, unless the context clearly dictates otherwise. The use of "first," "second," and similar terms in this disclosure is not intended to indicate any order, quantity, or importance, but rather is used to distinguish one element from another. Likewise, the word "comprising" or "comprises", and the like, means that the element or item listed before the word covers the element or item listed after the word and its equivalents, but does not exclude other elements or items. The terms "connected" or "coupled" and the like are not restricted to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

Furthermore, flow charts are used in this disclosure to illustrate operations performed by systems according to embodiments of the present disclosure. It should be understood that the preceding or following operations are not necessarily performed in the exact order in which they are performed. Rather, various steps may be processed in reverse order or simultaneously. Meanwhile, other operations may be added to the processes, or one or more operations may be removed from the processes.

Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

While the present disclosure has been described in detail above, it will be apparent to those skilled in the art that the present disclosure is not limited to the embodiments described in the present specification. The present disclosure can be implemented as modifications and variations without departing from the spirit and scope of the present disclosure defined by the claims. Accordingly, the description of the present specification is for the purpose of illustration and is not intended to be in any way limiting of the present disclosure.

Claims

1. A recommendation information processing method, comprising:

acquiring a recommendation information set containing a plurality of recommendation information;

extracting feature elements of a plurality of categories of each piece of recommendation information in the recommendation information set to obtain a feature element set; and

for each class of feature elements in the set of feature elements:

dividing a plurality of recommendation information in the recommendation information set into at least two groups, wherein attributes of feature elements of the category of recommendation information within each of the at least two groups are mutually exclusive from attributes of feature elements of the category of recommendation information within other of the at least two groups;

acquiring at least one result data corresponding to the recommendation information in each of the at least two groups;

for each of the at least one type of result data, a degree of difference between the result data of any two of the at least two groupings is calculated, and in the event that the degree of difference is greater than a first predetermined threshold, it is determined that the feature element of the category is significant to the result data.

2. The recommended information processing method according to claim 1, further comprising:

in the case where it is determined that the feature element of the category has significance to the result data, outputting information on the feature element of the category,

wherein the information comprises evaluation information indicating an influence of the feature elements of the category on the result data,

the method also includes generating a recommendation information analysis model for providing recommendation information production guidance based on the evaluation information.

3. The recommended information processing method according to claim 2, wherein the information further includes viewer characteristic information indicating a common characteristic of viewers of recommended information in a group of which result data is better among the two groups, the method further comprising:

and generating recommendation information oriented crowd recommendations based on the viewer characteristic information.

4. The recommended information processing method according to claim 2, further comprising:

and predicting the putting effect of the new recommendation information by extracting and analyzing the feature elements of a plurality of categories in the new recommendation information based on the output information about the feature elements of different categories.

5. The recommended information processing method according to claim 1, wherein the feature elements of the plurality of categories include feature elements of a plurality of basic attribute categories and/or a plurality of content attribute categories, wherein,

the characteristic elements of the plurality of basic attribute categories comprise at least one of frame rate, code rate, size, resolution, saturation, definition and lens number of recommendation information;

the characteristic elements of the plurality of content attribute categories include at least one of a shot, a scene, a person, an object, and a presentation form of the recommendation information.

6. The recommended information processing method of claim 5, wherein the at least two packets include a first packet and a second packet,

wherein dividing the plurality of recommendation information in the recommendation information set into at least two groups for each category of feature elements in the feature element set comprises:

for each recommendation information of the plurality of recommendation information, dividing the recommendation information into a first group if a value or a number of feature elements of the category of the recommendation information satisfies a predetermined condition; otherwise, the recommendation information is divided into a second group.

7. The recommendation information processing method of claim 5, wherein, for the feature elements of the plurality of content attribute categories, extracting the feature elements of the plurality of categories of each recommendation information in the recommendation information set comprises:

extracting feature elements of a plurality of content attribute categories in the recommendation information set using a deep learning model; and

for each recommendation information, generating that the recommendation information contains the confidence of the extracted feature element of each content attribute category and the confidence of the appearance position of the feature element of the content attribute category in the recommendation information.

8. The recommended information processing method according to claim 7, further comprising:

adjusting the output information about the feature elements of different content attribute categories by adjusting the confidence level.

9. The recommended information processing method according to claim 2, further comprising:

extracting a timestamp of each feature element in the feature element set; and

generating a recommendation information analysis model having time divisions for providing recommendation information production guidance based on the time stamps and the outputted information on the different categories of feature elements,

wherein the timestamp is appended to each feature element of the recommendation information when each recommendation information in the set of recommendation information is generated.

10. The recommendation information processing method of any one of claims 1-9, wherein the at least one outcome data comprises at least a portion of impression data, viewer behavior data, and industry distribution data, wherein,

the putting effect data comprises at least one item of recommended information click rate and recommended information conversion rate,

the viewer behavior data includes at least one of viewing duration, viewing times, daily repeat viewing times, play-out rate, jump-out rate, conversion time point, and conversion success rate of the recommended information viewer,

the industry distribution data indicates a distribution characteristic of an industry to which the recommendation information viewer belongs.

11. A recommendation information generation method includes:

extracting characteristic elements of a plurality of categories in the recommendation information to be optimized; and

for each of the plurality of categories of feature elements:

dividing the recommendation information to be optimized into at least two groups, wherein the attribute of the characteristic element of the category of recommendation information in each of the at least two groups is mutually exclusive with the attribute of the characteristic element of the category of recommendation information in the other of the at least two groups;

predicting at least one result data of the recommendation information to be optimized according to predetermined evaluation information about the grouping of the characteristic elements aiming at the category based on the grouping where the recommendation information to be optimized is located;

optimizing feature elements of the category of the recommendation information to be optimized by using the at least one predicted result data.

12. A recommended information processing apparatus comprising:

a recommendation information acquisition unit configured to acquire a recommendation information set containing a plurality of recommendation information;

a feature extraction unit configured to extract feature elements of a plurality of categories of each recommendation information in the recommendation information set to obtain a feature element set; and

a recommendation information processing unit configured to, for each category of feature elements in the feature element set:

13. A recommendation information processing apparatus comprising:

one or more processors; and

one or more memories, wherein the memory has stored therein computer-readable code, which, when executed by the one or more processors, causes the one or more processors to perform the method of any one of claims 1-11.

14. A computer readable storage medium having computer readable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method of any one of claims 1-11.

15. A computer program product comprising computer readable instructions which, when executed by a processor, cause the processor to perform the method of any one of claims 1-11.