Short video supply and demand diagnosis method and device

Abstract

The present application relates to the technical field of video recommendation, and specifically provides a short video supply and demand diagnosis method and device, which is applied to short video recommendation, constructs a short video replenishment model through the average short video play time of a short video platform and the video play quantity of a short video, and builds the short video replenishment model as follows: S1, collecting supply and demand data on the short video platform, calculating the supply and demand ratio based on the collected supply and demand data, judging the content value gap according to the supply and demand ratio, and taking the out-of-stock time length as the measurement of the content value gap; S2, calculating the video production value using the marginal distribution value, wherein the content value measurement is that each additional or produced short video content can bring time length increment to the short video market; S3, obtaining the optimal short video replenishment scheme according to the out-of-stock time length and the video production value. Compared with the prior art, the present application can provide a reference for the content acquisition of the platform, improve the overall ROI of content acquisition, and make the short video platform reach supply and demand balance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of video recommendation, and specifically provides a short video supply and demand diagnosis method and device. BACKGROUND

[0002] At present, most short video businesses do not have a practical and effective and systematic solution to determine whether the supply and demand status is problematic. A small number of short video businesses may determine whether the traffic is too concentrated in a specific category based on artificial experience. Some individual businesses may explore the supply and demand of different categories by conducting a large number of random experiments and giving different categories of weight.

[0003] The method of artificial experience judgment has the problems of inaccurate artificial judgment and limited application scenarios. On the one hand, artificial judgment of whether the content of a single category is too large or too small is mainly subjective judgment by short video depth users or experts, but the judgment standard cannot be accurately described or quantified, and the standard cannot be consistent in actual execution. On the other hand, artificial experience is limited in application scenarios. When the problem is very obvious, artificial judgment can be made on whether a single category is too much or too little. For example, when the game category accounts for 80% of the traffic, the ecology is obviously unhealthy. However, when the problem is not so obvious, such as when all categories evenly divide the traffic, artificial judgment cannot determine whether the current state is healthy or not, or which direction the ecology should adjust to, or how much of a specific category should be introduced.

[0004] If a large number of random experiments are conducted, there is a problem of large traffic occupation and high experimental cost. Indeed, the distribution of specific categories can be tried to explore whether a specific category is in a state of shortage or redundancy by increasing or decreasing the weight. However, there are many categories, and it is not known whether the weight should be increased or decreased. If a stable result is to be obtained, many experiments need to be conducted, and these experiments do not always have a positive expected benefit, which is a great waste of traffic.

[0005] There is also a problem of being unable to determine whether it is globally optimal, because a good result is obtained based on a certain experiment, but this only verifies whether the experimental strategy is better than the control group, and it is completely unknown whether there is further improvement space, which cannot systematically guide the business.

[0006] Therefore, how to determine which categories of content need to be supplemented to achieve supply and demand balance while minimizing the cost of replenishment is a problem to be solved by the present application. SUMMARY

[0007] The technical task of the present application is to provide a short video supply and demand diagnosis method and device to solve the problem of judging which categories of content need to be supplemented additionally, enabling the platform to achieve supply and demand balance while minimizing the cost of replenishment.

[0008] The technical task of the present application is achieved in the following way: a short video supply and demand diagnosis method applied to short video recommendation, a short video replenishment model is constructed by the average short video play time of the short video platform and the video play volume of the short video, and the short video replenishment model is constructed as follows:

[0009] S1, collecting supply and demand time length data on the short video platform, calculating the supply and demand ratio based on the collected supply and demand time length data, judging the content value gap according to the supply and demand ratio, and taking the out-of-stock time length as the measurement of the content value gap;

[0010] S2, calculating the video production value using the marginal distribution value, the content value measurement being the time length increment that can be brought to the short video market for each additional short video content introduced or produced;

[0011] S3, obtaining the optimal short video replenishment scheme according to the out-of-stock time length and the video production value.

[0012] Further, in step S1, the supply and demand measurement index is the supply and demand ratio, and the supply and demand ratio is:

[0013]

[0014] The supply and demand ratio is a scalar between 0 and 1, and the closer to 1, the more balanced the supply and demand of the short video platform; the closer to 0, the more serious the supply problem of the short video platform;

[0015] Wherein, the supply time length is equal to the actual consumption time length, i.e. the time length actually met by the short video platform under the current content supply and content distribution;

[0016] The demand time length is equal to the maximum consumable time length, i.e. considering that there is no restriction on content supply and no restriction on content distribution, the upper limit of the time length that the current user can consume on the short video platform.

[0017] Further, the out-of-stock time length is the migration of the user group in the low consumption time length to the higher user group by adjusting the supply and demand structure, wherein,

[0018] Out-of-stock time length = (average time length of high quantile population - average consumption time length) * size of users below high quantile population.

[0019] Further, in step S2, the video production value is P, and the marginal distribution value is S,

[0020] P = (Δ (time length) | Δ (number of replenishment contents) ) = Δ (number of replenishment contents) S; wherein S = (Δ (time length) | do (number of distribution times) ) ;

[0021] The is a pool funnel, the is a recommended distribution preference, the Δ (number of replenishment contents) is a content production change, and the S is a time length increment that a video play per additional distribution of a short video can bring to a short video market.

[0022] Further, the pool funnel is that when creators create short video contents, the short video contents are audited by humans or machines, and only high-quality short video contents that meet the conditions are put into the recommendation pool;

[0023] The recommended distribution preference is to expose different contents based on user preference estimation under the premise of determining the recommendation pool.

[0024] The pool funnel and the recommended distribution preference are considered as constants.

[0025] Further, when calculating the marginal distribution value S, the play order of the video needs to be considered. Let T be the number of effective short video plays in a certain category, Y be the total play time of subsequent videos brought by effective plays, and X be the covariate to be controlled.

[0026] When the result variable Y runs, first, define the short video played in a certain time period as the number of effective short video plays, and eliminate the short video play behavior of users who quickly slide due to lack of interest. Then, evenly distribute the play time of each short video to the number of effective short video plays, and the total time brought by each effective short video play is the result variable Y.

[0027] Further, in step S3, the optimal short video replenishment scheme is:

[0028] ;

[0029] The is a replenishment cost measure, the is a distribution difference measure of the category distribution of short video replenishment and the historical consumption habits of users, and the product of the replenishment cost measure and the distribution difference measure of the category distribution of short video replenishment and the historical consumption habits of users is used as an optimization target. The smaller the optimization target is, the better.

[0030] A short video supply and demand diagnosis device, comprising at least one memory and at least one processor.

[0031] ​The at least one memory is configured to store a machine-readable program.

[0032] The at least one processor is configured to invoke the machine-readable program to execute a short video supply and demand diagnosis method.

[0033] The short video supply and demand diagnosis method and device have the following advantages:

[0034] The present application not only provides a measurement of the supply and demand status, but also calculates the duration gap of the short video market, the content value measurement, and a cost-optimal short video replenishment scheme, which is expected to provide a reference for content acquisition of short video platforms and improve the overall ROI of content acquisition, and also enable the short video platform to achieve supply and demand balance. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0036] FIG. 1 is a flowchart of a short video supply and demand diagnosis method; Figure 1 FIG. 1 is a flowchart of a short video supply and demand diagnosis method;

[0037] FIG. 2 is a schematic diagram of calculating marginal distribution value in a short video supply and demand diagnosis method; Figure 2 FIG. 2 is a schematic diagram of calculating marginal distribution value in a short video supply and demand diagnosis method; DETAILED DESCRIPTION

[0038] The voice recognition-based retail terminal human-computer interaction method, system, device and medium of the present application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] Embodiment 1:

[0040] As shown in FIG. 1, the present embodiment provides a short video supply and demand diagnosis method, which specifically includes the following steps: Figure 1 As shown in FIG. 1, the present embodiment provides a short video supply and demand diagnosis method, which specifically includes the following steps:

[0041] Applied to short video recommendation, a short video replenishment model is constructed by using the average short video play time of the short video platform and the video play time of the short video, and the short video supply and demand is diagnosed using the short video replenishment model to give the optimal short video replenishment scheme.

[0042] The short video replenishment model is constructed as follows:

[0043] S1, collecting supply and demand time data on the short video platform, calculating the supply and demand ratio based on the collected supply and demand time data, judging the content value gap according to the supply and demand ratio, and taking the out-of-stock time as a measurement of the content value gap;

[0044] Among them, the supply-demand measurement indicator is the supply-demand ratio, which is:

[0045]

[0046] The supply-demand ratio is a scalar between 0 and 1. The closer it is to 1, the more balanced the supply and demand of short video platforms are; the closer it is to 0, the more serious the supply problem of short video platforms is.

[0047] Supply duration equals actual consumption duration, which is the actual time that a short video platform satisfies the current user under the current content supply and distribution conditions;

[0048] Demand duration equals maximum consumable duration, which is the upper limit of the time a user can consume on a short video platform under the condition that there are no restrictions on content supply or distribution.

[0049] To increase supply duration, the following two aspects can be considered:

[0050] (1) When the content supply of the platform is stable, improving the distribution mechanism and increasing the distribution efficiency will allow the existing content of the platform to reach users who are interested in it, thereby increasing user consumption and directly improving the supply-demand ratio.

[0051] (2) When the content distribution is consistent, if the new content replenishment is of interest to users, it can also increase users’ consumption and improve the supply-demand ratio.

[0052] like:

[0053] By identifying any supply unit and demand unit pair, we can calculate the supply-demand ratio for this pair.

[0054] Supply Unit: Videos of the same category can be regarded as a supply unit. On this basis, more features can be added to further refine the supply unit. For example, content of the same category with a video length of less than 5 minutes and an author with more than 100,000 followers can also be regarded as the same supply unit. The core assumption of the supply unit is that the content within the same supply unit is homogeneous. This can help us abstract hundreds of millions of different short videos into dozens or hundreds of supply units.

[0055] Demand Unit: Users of the same age, gender, and region can be considered as a demand unit. Similarly, constraints such as user activity level and interest preferences can be added to further refine the demand unit. The core assumption of demand units is that users in the same demand unit are homogeneous.

[0056] Supply-demand ratio estimation: In the supply-demand ratio estimation, the supply duration of the short video platform can be directly counted through historical data, and the demand duration of the short video platform needs to be estimated by selecting a suitable method. We believe that relatively active users (such as the top 90 percentile of average duration) can be considered as the upper limit of single-user demand duration. By improving the content ecosystem, the activity level of relatively inactive users can be improved to the 90th percentile. The specific percentile selection can be further adjusted according to the business development stage.

[0057] With the supply-demand ratio, further calculate the supply duration, and the shortage duration is to migrate the user group in the low consumption duration to the higher percentile user group by adjusting the supply structure, wherein,

[0058] Shortage duration = (average duration of high percentile users - average consumption duration) * size of users below high percentile.

[0059] If the high percentile is 90th percentile, then:

[0060] Shortage duration = (average duration of high percentile users - average consumption duration) * size of users below high percentile.

[0061] S2, calculate the video production value using the marginal distribution value, which is the content value metric that each additional short video content introduced or produced can bring time increment to the short video market;

[0062] Assuming the video production value is P, and the marginal distribution value is S,

[0063] P = (Δ (duration) | Δ (content quantity)) = Δ (content quantity) * * S; wherein S = (Δ (duration) | do (distribution times));

[0064] Wherein, is the funnel, is the recommendation distribution preference, Δ (content quantity) is the content production change, S is that when the platform distributes one more content, the user feels satisfied after seeing the corresponding content, which will generate more content browsing behavior, thereby bringing more consumption duration. The marginal distribution value refers to how much time increment each additional short video play can bring to the market.

[0065] Funnel is when creators create short video content, short video content is audited by humans or machines, and only high-quality short video content that meets the conditions will enter the recommendation pool;

[0066] Recommendation distribution preference is that under the premise of determining the recommendation pool, based on the estimation of user preferences, different contents are exposed to different users.

[0067] Where the funnel into the pool and the recommended distribution preferences are difficult to influence from the perspective of replenishment, they can be regarded as constants; while replenishment will indirectly make the platform distribute more short videos, thus affecting the marginal distribution value of the content, which needs to be solved.

[0068] Because users will continue to brush short videos only after they see the content and feel like it, that is, the video play of the previous short video will bring the subsequent duration, so the video play order needs to be considered in the modeling process.

[0069] Assume that the intervention variable T is the number of video plays of effective short videos in a certain category; the result variable Y is the sum of the subsequent video play duration brought by effective play; and the control variable X is the covariate that needs to be controlled, such as the user's age, gender, region, activity level, historical consumption duration of a certain category, video play of short videos, and other features.

[0070] In this embodiment, the causal inference model Double Machine Learning is used to solve it, which needs to control the covariate unchanged, and get the result variable Y through the intervention variable T.

[0071] Where, the result variable Y is: because users will continue to brush after seeing a video and feeling satisfied, the subsequent video play behavior can be considered as the result of the joint action of each video play behavior, and the video play duration of each video is evenly divided into the video play of the previous effective short video. In this embodiment, the definition of the video play of the effective short video is to play a video for more than three seconds, so as to eliminate the video play behavior of quickly sliding down due to lack of interest, and the sum of the duration brought by the video play of each effective short video is the result variable Y.

[0072] As shown in Figure 2 , the examples are five short videos in turn, the video play duration of the first short video is 10S, because the video play of more than three seconds is defined as the effective short video play, so the first short video is the effective short video play, and the video play duration is 10S / 1;

[0073] The video play duration of the second short video is 2S, which is less than 3S, so it is invalid short video play and does not bring additional duration, and its play duration is brought by the first short video, so the play of 2S is added to the first short video play, which is represented as 2S / 1;

[0074] The video play duration of the third short video is 13S, which is greater than 3S, so it is effective short video play, which is brought by the first and third short videos, and the played 13S is added to the first and third short video plays respectively, which is represented as 13S / 2;

[0075] The video playing time length of the fourth short video is 15S, which is greater than 3S, so it is an effective short video playing, which is caused by the first, third and fourth short video playing, so 15S is added to the first, third and fourth short video playing respectively, which is expressed as 13S / 3;

[0076] The video playing time length of the fifth short video is 2S, which is less than 3S, so it is an invalid short video playing, which is caused by the first, third and fourth short video playing, so 2S is added to the first, third and fourth short video playing respectively, which is expressed as 2S / 3;

[0077] Therefore, the result variable Y in the example is 10S / 1+2S / 1+13S / 2+15S / 3+2S / 3.

[0078] S3, according to the out-of-stock time length and the video production value, the optimal short video replenishment scheme is obtained.

[0079] The optimal short video replenishment scheme is:

[0080] ;

[0081] The replenishment cost is measured as: The category distribution of short video replenishment and the distribution difference of user historical consumption habits are measured, and the replenishment cost and the category distribution of short video replenishment and the distribution difference of user historical consumption habits are multiplied as the optimization target, and the smaller the optimization target is, the better, in order to prevent the content distribution from changing dramatically because of replenishment, thereby affecting the user experience.

[0082] The above short video supply and demand diagnosis and short video replenishment model can refer to the following steps when implemented:

[0083] All user consumption browsing time lengths on short video products within a day are collected through user operation of the client, and the corresponding data is stored in a big data storage engine, such as hive, mongodb, redis, etc. The original user behavior log (such as consumption time length, click rate), user portrait (age, gender, region) and content metadata are stored.

[0084] The client can include a smart phone, a desktop computer, a tablet computer, a notebook computer, a smart device, etc.

[0085] Using the big data operating system inside the short video platform, relevant data is processed from the database through sql or pyspark language, including data cleaning, filtering invalid data (such as abnormal short-time clicks), anonymizing and encrypting user privacy data (such as region, age), to obtain supply and demand duration data, and according to the supply and demand duration data, the supply and demand ratio and the out-of-stock duration are constructed.

[0086] And through sql or python language, the user's age, gender, region, consumption and activity data in the past 30 days are processed into modeling features from the database, and modeling and solving are performed on windows or macOS systems through python, to obtain the marginal distribution value S of the content, and the actual index value of the pool funnel and recommendation preference parameters is counted from the database, so as to obtain the video production value P, and then through python, the content replenishment scheme is obtained by operational planning and solving;

[0087] The business operator will purchase and replenish content based on the replenishment scheme, input the purchased and replenished content into the storage of the short video business background at one time, use Ab test to compare the user satisfaction before and after replenishment, dynamically adjust the model parameters, and finally present to the client through the links such as pool funnel and recommendation preference.

[0088] Embodiment 2:

[0089] The embodiment also provides a short video supply and demand diagnosis device, comprising at least one memory and at least one processor.

[0090] The memory is used to store machine readable programs.

[0091] The processor is used to call the machine readable programs and execute a short video supply and demand diagnosis method.

[0092] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), ready-to-program gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor can be a microprocessor, or the processor can be any conventional processor.

[0093] The memory can be used to store computer programs and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, etc.; and the data storage area can store data created according to the use of the terminal, etc. In addition, the memory can also include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a secure digital (SD) card, a flash card, at least one disk storage period, a flash memory device, or other volatile solid-state memory devices.

[0094] The above specific embodiments are only specific cases of the present application, and the patent protection scope of the present application includes but is not limited to the above specific embodiments. Any technical solution meeting the above specific embodiments of the present application and any appropriate changes or replacements made by those skilled in the art to the technical solution shall fall within the patent protection scope of the present application.

[0095] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1.A short video supply and demand diagnosis method, characterized in that, The application is applied to short video recommendation. A short video replenishment model is constructed by using the average short video play time of users in a short video platform and the video play volume of short videos. The short video replenishment model is constructed as follows: S1. Collecting supply-demand time length data in the short video platform, calculating the supply-demand ratio based on the collected supply-demand time length data, judging the content value gap according to the supply-demand ratio, and taking the shortage time length as the measurement of the content value gap; The supply-demand ratio is: ； The supply-demand ratio is a scalar between 0 and 1. The closer to 1, the more balanced the supply and demand of the short video platform. The closer to 0, the more serious the supply problem of the short video platform; Wherein, the supply time length is equal to the actual consumption time length, that is, the time length actually met by the current user under the current content supply and content distribution of the short video platform; The demand time length is equal to the maximum consumable time length, that is, considering that there is no any content supply restriction and no any content distribution restriction, the upper limit of the time length that the current user can consume in the short video platform; The shortage time length is to migrate the user group in the low consumption time length to the higher user group by adjusting the supply-demand structure, wherein, Shortage time length = (average time length of high quantile population - average consumption time length) * user scale below high quantile population; S2. Calculate the video production value using the marginal distribution value. The content value measurement is the time length increment that can be brought to the short video market by introducing or producing one more short video content; The video production value is P, and the marginal distribution value is S, P = (Δ (duration) | Δ (number of replenishment contents)) = Δ (number of replenishment contents) ; wherein S = (Δ (duration) | do (number of distributions)) P = (Δ (duration) | Δ (number of replenishment contents)) = Δ (number of replenishment contents) ; wherein S = (Δ (duration) | do (number of distributions)) P = (Δ (duration) | Δ (number of The As a funnel into the pool, the As a recommended distribution preference, the delta (number of replenishment content) is a content production change, and S is the time increment that each additional distribution of a short video can bring to the short video market; The funnel into the pool is that when the creators create short video content, the short video content is audited by artificial or machine, and only the high-quality short video content meeting the conditions can enter the recommendation pool; The recommendation distribution preference is to expose different contents based on the user preference estimation under the premise of determining the recommendation pool; The funnel into the pool and the recommendation distribution preference are considered as constants; When calculating the marginal distribution value S, the play order of the video needs to be considered. Let T be the video play volume of the effective short video under a certain category, Y be the time length sum of the subsequent video play brought by the effective play, and X be the control variable that needs to be controlled. Double Machine Learning is used to solve the problem. In this model, the control variable needs to be constant, and the result variable Y is obtained through the intervention variable T. Wherein, the result variable Y is: because the user sees a video and feels satisfied, he will continue to scroll down, so the subsequent video play behavior can be considered as the result of the joint action of each previous video play behavior. The video play time of each video is evenly divided into the video play volume of the previous effective short video. The definition of the video play volume of the effective short video is that the video is played for more than three seconds, so as to eliminate the video play behavior of the user quickly sliding down due to lack of interest. The time length sum brought by the video play volume of each effective short video is the result variable Y; S3. Obtain the optimal short video replenishment scheme according to the shortage time length and the video production value; In step S3, the optimal short video replenishment scheme is: ； The As a replenishment cost metric, the As a distribution difference metric of the category distribution of short video replenishment and the distribution of user historical consumption habits, the product of the replenishment cost metric and the distribution difference metric of the category distribution of short video replenishment and the distribution of user historical consumption habits is taken as an optimization target, and the smaller the optimization target is, the better. 2.A short video supply and demand diagnosis apparatus, characterized in that, It includes: At least one memory and at least one processor; The at least one memory is configured to store a machine readable program; The at least one processor is configured to invoke the machine readable program to execute the method of claim 1.

Images