A video data recommendation method and system based on label analysis

Abstract

The present invention discloses a video data recommendation method and system based on label analysis, which relates to the technical field of video recommendation, including obtaining historical viewing information, calculating and analyzing the viewing rate; performing statistics on each type of label information, and screening and processing the statistical results; using a large language model to analyze label combinations, querying historical viewing information, and making contingency judgments on label combinations that combine effective information to recommend videos to users; the present invention is used to solve the problem that in the prior art, due to the lack of analysis of video labels of user viewing records and the lack of an easy-to-implement video recommendation method, when a poor video recommendation model obtained during training is used to recommend videos to users, the recommended content may not match the user's interests, resulting in a poor user viewing experience.

Description

Technical Field

[0001] The present invention relates to the technical field of video recommendation, and in particular to a video data recommendation method and system based on tag analysis. Background Art

[0002] The amount of short video content in the industry is usually in the millions, and some professional video recommendation applications even have tens of millions of video content. It is obviously not realistic to directly sort all of this content. At this time, video recommendation technology is needed to screen the content of the video library, while ensuring that the video content is close to the user's interests, reducing the amount of content involved in the sorting.

[0003] Existing improvements for video recommendation usually use a large amount of historical viewing data to train a video recommendation model, and then use the trained video recommendation model to recommend videos to users. For example, in a patent application with invention publication number CN117669687A, a video recommendation model training method, a video recommendation method, a device and a storage medium are disclosed. The solution is to pre-train the initial video recommendation model by using multiple unlabeled video samples to obtain a pre-trained video recommendation model; then, according to the user's feedback behavior on the recommended video, the parameters of the video recommendation model are adjusted until the training end conditions are met to obtain a trained video recommendation model; in the method of training the video recommendation model, the quality of the obtained model depends on the number of training samples and the internal structure of the model, and the training process cannot be intervened. There is a lack of analysis of the video tags of the user's viewing records, and a transparent and easy-to-implement video recommendation method is provided. In view of this, it is necessary to improve the existing video recommendation method. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems in the prior art to a certain extent. By improving the video recommendation method, it is used to solve the problem that the prior art lacks analysis of video tags of user viewing records and lacks an easy-to-implement video recommendation method. When a poor video recommendation model obtained during the training process is used to recommend videos to users, the recommended content may not match the user's interests, resulting in a poor user viewing experience.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a video data recommendation method based on tag analysis, comprising:

[0006] Acquire a first amount of historical viewing information, the historical viewing information including duration information and tag information; calculate the duration information and output a viewing rate; calculate and analyze the viewing rate, and output an effective rate and an effective ratio based on the calculation and analysis results;

[0007] Based on the effective ratio, historical viewing information is acquired for a second time, and the historical viewing information is filtered and processed based on the effective rate;

[0008] Perform statistics on each type of tag information, calculate the statistical results, and output the tag ratio; perform calculation and analysis on the tag ratio, output the low-frequency ratio, and filter the statistical results based on the low-frequency ratio;

[0009] The filtered label information is combined to obtain multiple label combinations; the label combinations are analyzed using a large language model, and the label combinations are deleted or not processed based on the analysis results; historical viewing information is queried based on the label combinations, the query results are analyzed, and valid combination information and invalid combination information are output based on the analysis results;

[0010] Perform a random judgment on the label combination of combined effective information, and output the combined random information or the combined common information based on the random judgment result; and give priority to recommending videos to users based on the label combination of combined common information.

[0011] Further, the duration information includes video duration and viewing duration; obtaining a first amount of historical viewing information, calculating the duration information, and outputting a viewing rate; calculating and analyzing the viewing rate, and outputting an effective rate and an effective proportion based on the calculation and analysis results, including:

[0012] Obtain a first amount of historical viewing information, calculate a ratio of viewing time of each piece of historical viewing information to video time, and mark the ratio as viewing rate;

[0013] Execute an effective calculation cycle, and output the effective rate and effective ratio based on the effective calculation cycle.

[0014] Furthermore, the effective calculation cycle includes:

[0015] Set the reference rate with an initial value of 100% and the effective number with an initial value of 0;

[0016] Compare each viewing rate with the reference rate. When the viewing rate is less than the reference rate, no processing is done; when the viewing rate is greater than or equal to the reference rate, the effective number is increased by one.

[0017] Calculate the ratio of the significant number to the first number, labeled as the analysis ratio;

[0018] When the analysis rate is greater than or equal to the first frequency, the reference rate is marked as effective, the analysis rate is marked as effective proportion, and the effective calculation cycle ends;

[0019] When the analysis rate is less than the first frequency, the effective number is reset to 0, the value obtained by subtracting the second frequency from the reference rate is assigned to the reference rate, and the effective calculation cycle is executed again.

[0020] Further, the historical viewing information is acquired for a second time based on the effective ratio, and the historical viewing information is filtered based on the effective ratio, including:

[0021] Calculate the product of the first quantity and the effective proportion, and mark it as the first effective number;

[0022] Calculate the difference between the second number and the first effective number, marking it as the required number;

[0023] Calculate the quotient of the required number and the effective proportion, and mark it as the number to be obtained again;

[0024] Obtain the historical viewing information of the number of re-obtained items, and calculate the viewing rate of each historical viewing information;

[0025] The historical viewing information with a viewing rate lower than the effective rate will be deleted.

[0026] Further, statistics are collected for each type of tag information, the statistical results are calculated, and the tag proportion is output; the tag proportion is calculated and analyzed, the low-frequency proportion is output, and the statistical results are screened based on the low-frequency proportion, including:

[0027] Get the number of remaining historical viewing information, marked as the number of samples;

[0028] Count each tag information that appears in all historical viewing information, and mark the count result as the occurrence frequency;

[0029] Calculate the ratio of each occurrence frequency to the number of samples, and mark it as the label ratio;

[0030] The first scaling multiple of the largest label ratio is marked as the low-frequency ratio, and label information corresponding to the label ratio that is smaller than the low-frequency ratio is deleted.

[0031] Further, the filtered label information is combined to obtain multiple label combinations; the label combinations are analyzed using a large language model, and the label combinations are deleted or not processed based on the analysis results, including:

[0032] Combine the remaining label information in pairs so that each combination exists, and obtain multiple label combinations;

[0033] Use the large language model to analyze the two tag information in the tag combination to determine whether the two tag information are included in the relationship or point to the same meaning;

[0034] When the judgment result is yes, the label combination is deleted; when the judgment result is no, no processing is performed.

[0035] Further, querying the historical viewing information based on the tag combination, analyzing the query result, and outputting the combination valid information and the combination invalid information based on the analysis result include:

[0036] Marking two tag information in the tag combination as a first tag and a second tag respectively; querying the historical viewing information using the first tag as an index;

[0037] Mark all the historical viewing information obtained by querying with 1;

[0038] Query the historical viewing information using the second tag as an index;

[0039] All the historical viewing information obtained by query is marked as 2;

[0040] The historical viewing information is analyzed. If there is no historical viewing information marked with both 1 and 2, the combination invalid information is output; if there is, the historical viewing information is marked as combined viewing information, and the combination valid information is output.

[0041] Furthermore, the combination of labels of combined valid information is subjected to accidental judgment, and combined accidental information or combined common information is output based on the accidental judgment result, including:

[0042] The combination of tags that combine valid information is subjected to a contingency judgment, wherein the contingency judgment includes:

[0043] Mark the largest occurrence frequency in the tag combination as the main frequency, count the total number of combined viewing information corresponding to the tag combination, and mark it as the analysis frequency;

[0044] Calculate the ratio of the analysis frequency to the main frequency, marked as the frequency ratio;

[0045] The first tag and the second tag in the tag combination are used as indexes for querying, and the difference between the upload times of two adjacent query results is calculated and marked as the time interval;

[0046] Use the accidental analysis formula to calculate the frequency ratio and time interval, and output the accidental judgment value;

[0047] The accidental analysis formula is configured as: ; Where Ro is the accidental judgment value, Pz is the frequency ratio value, Tz is the time interval, and k is a constant;

[0048] The accidental analysis value is analyzed. If the accidental analysis value is greater than or equal to the accidental threshold, the combined common information is output; if the accidental analysis value is less than the accidental threshold, the combined accidental information is output.

[0049] Furthermore, the tag combinations based on the combination of common information are used to prioritize videos recommended to users, including:

[0050] Mark the tag combinations that combine common information as priority combinations;

[0051] Analyze the video information stored in the cloud in real time, and when the stored video information includes two tag combinations in the priority combination at the same time, mark the video information as a priority video;

[0052] Push priority videos to users first.

[0053] In a second aspect, the present invention also provides a video data recommendation system based on tag analysis, including a habit analysis module, a tag statistics module, a tag analysis module and a video push module;

[0054] The habit analysis module is used to obtain a first amount of historical viewing information, wherein the historical viewing information includes duration information and tag information; calculate the duration information and output the viewing rate; calculate and analyze the viewing rate, and output the effective rate and effective ratio based on the calculation and analysis results;

[0055] The habit analysis module is also used to obtain historical viewing information for a second time based on the effective ratio, and filter and process the historical viewing information based on the effective ratio;

[0056] The tag statistics module is used to count each tag information, calculate the statistical results, and output the tag proportion; calculate and analyze the tag proportion, output the low-frequency proportion, and filter the statistical results based on the low-frequency proportion;

[0057] The tag analysis module is used to combine the tag information after screening to obtain multiple tag combinations; use the large language model to analyze the tag combinations, and delete or do not process the tag combinations based on the analysis results; query the historical viewing information based on the tag combinations, analyze the query results, and output combination valid information and combination invalid information based on the analysis results;

[0058] The label analysis module is also used to make a contingency judgment on the label combination of the combined valid information, and output the combined contingency information or the combined common information based on the contingency judgment result;

[0059] The video push module is used to preferentially recommend videos to users based on tag combinations that combine common information.

[0060] Beneficial effects of the present invention: The present invention obtains a first amount of historical viewing information and performs calculation and analysis to output the efficiency and the effective proportion, obtains the historical viewing information for a second time based on the effective proportion, and screens the historical viewing information based on the efficiency. This has the advantage of being able to obtain a relatively accurate amount of analysis data as needed, avoiding the situation in which too much invalid data is obtained once, resulting in unclear secondary acquisition, and reducing the calculation efficiency of the system, thereby improving the efficiency and accuracy of habit analysis;

[0061] The present invention also analyzes and screens the labels of the videos, combines the remaining label information, performs rationality analysis and contingency analysis on the combinations in turn, obtains a final priority combination based on the analysis results, and finally recommends videos that include two labels in the priority combination to the user based on the priority combination; the advantage of this is that through rationality analysis and contingency analysis, multiple labels that the user is interested in can be accurately combined together, the user pays more attention to them, the user's viewing experience can be improved, and the intelligence and accuracy of video push are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a functional block diagram of the system of the present invention.

[0063] Figure 2 The figure is a flow chart of the steps of the method of the present invention.

[0064] Figure 3 It is a partial data flow diagram of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] Example 1, first aspect, please refer to Figure 1 as well as Figure 3 As shown, the present invention provides a video data recommendation system based on tag analysis, including a habit analysis module, a tag statistics module, a tag analysis module and a video push module;

[0067] The habit analysis module is used to obtain a first amount of historical viewing information, the historical viewing information including duration information and tag information; calculate the duration information and output the viewing rate; calculate and analyze the viewing rate, and output the effective rate and effective ratio based on the calculation and analysis results;

[0068] The habit analysis module is also used to obtain historical viewing information for a second time based on the effective ratio, and filter and process the historical viewing information based on the effective ratio;

[0069] The habit analysis module is configured with a viewing analysis strategy, which includes:

[0070] Obtain a first amount of historical viewing information, calculate a ratio of viewing time of each piece of historical viewing information to video time, and mark the ratio as viewing rate;

[0071] In a specific implementation, the first number is set to 100, and the first number of historical viewing information is used to predict the user's video viewing habits; for example, the user is accustomed to quit watching or watch other videos when the video is 70% played, and a small but appropriate amount of historical viewing information is needed to predict the user's viewing habits. Therefore, the first number is set to 100, which reduces the amount of data for calculation while ensuring that the obtained data has sufficient data support;

[0072] Execute an effective calculation cycle, and output the effective rate and effective ratio based on the effective calculation cycle.

[0073] The effective calculation loop includes: setting a reference rate with an initial value of 100% and an effective number with an initial value of 0; comparing each viewing rate with the reference rate, and when the viewing rate is less than the reference rate, no processing is performed; when the viewing rate is greater than or equal to the reference rate, adding one to the effective number; calculating a ratio of the effective number to the first number, marking it as an analysis rate; when the analysis rate is greater than or equal to the first frequency, marking the reference rate as an effective rate, marking the analysis rate as an effective proportion, and ending the effective calculation loop; when the analysis rate is less than the first frequency, resetting the effective number to 0, assigning the value obtained by subtracting the second frequency from the reference rate to the reference rate, and executing the effective calculation loop again;

[0074] In the specific implementation, the first frequency is set to 50% and the second frequency is set to 10%; the first frequency is to simulate the situation that the user is interested in half of the videos but not interested in the other half of the data. When the user is interested in more than half of the videos, the efficiency can be obtained in advance and the effective calculation cycle can be ended; on the other hand, based on the existing video recommendation technology, there is only a small probability that more than 50 videos in 100 historical viewing data are not of interest to the user. Combining these two aspects, the first frequency is set to 50%; the second frequency is set to 10% in order to reduce the number of cycles while improving the reliability of the obtained efficiency.

[0075] The habit analysis module is also configured with a data supplement strategy, which includes: calculating the product of the first number and the effective proportion, marking it as the first effective number; calculating the difference between the second number and the first effective number, marking it as the required number; calculating the quotient of the required number and the effective proportion, marking it as the number to be obtained again; in specific implementation, the second number is set to 500; after obtaining the historical viewing information for the first time and calculating and analyzing the effective proportion, if the effective proportion is 60%, that is, 60 videos out of every 100 historical viewing information are videos of interest to the user, then the first effective number is 60, and the required number is calculated to be 440, and the number to be obtained again is approximately equal to 734, which means that after obtaining 734 historical viewing information again, 500 historical viewing data with reference significance can be left for analysis after screening;

[0076] Compared with the first number being set to 100, the efficiency of the calculation and analysis needs to be considered when setting the first number, and the second number being set to 500 is intended to accurately analyze and obtain tag information that users are interested in in the short term. If it is also necessary to analyze tags of users' long-term interest, the second number can be increased accordingly.

[0077] Obtain the historical viewing information of the number of times the video is obtained again, and calculate the viewing rate of each historical viewing information; delete the historical viewing information with a viewing rate lower than the effective rate; it should be noted that, taking the effective rate of 60% as an example, for any video, if the user watches the video for a time period greater than or equal to 60% of the total length of the video, it indicates that the user is interested in the video, and at this time, the label information corresponding to the video has analytical significance; and when the user watches the video for a time period less than 60% of the total length of the video, it may indicate that the user is not interested in the video directly or that the user stops watching the video after watching it for a short time, which can also indirectly prove that the user is not interested in the video or is not very interested in the video, and at this time, the label information corresponding to the video has no analytical significance, and therefore is deleted.

[0078] The tag statistics module is used to count each tag information, calculate the statistical results, and output the tag ratio; calculate and analyze the tag ratio, output the low-frequency ratio, and filter the statistical results based on the low-frequency ratio;

[0079] The tag statistics module is configured with a tag statistics strategy, which includes: obtaining the number of remaining historical viewing information, marking it as the number of samples; counting each tag information appearing in all historical viewing information, and marking the counting result as the frequency of occurrence; calculating the ratio of each frequency of occurrence to the number of samples, marking it as the tag proportion; marking the first zoom multiple of the largest tag proportion as the low-frequency proportion, and deleting the tag information corresponding to the tag proportion that is less than the low-frequency proportion; in specific implementation, the first zoom multiple is set to 0.5 times. When the tag proportion is less than 0.5 times the largest tag proportion, it means that the user is not very interested in the tag information, and it is not worthy of giving priority to recommending videos related to the tag information, and it is directly deleted in the tag statistics strategy.

[0080] The tag analysis module is used to combine the tag information after screening to obtain multiple tag combinations; use the large language model to analyze the tag combination, and delete or do not process the tag combination based on the analysis results; query the historical viewing information based on the tag combination, analyze the query results, and output the combination valid information and combination invalid information based on the analysis results;

[0081] The label analysis module is also used to make a contingency judgment on the label combination of the combined valid information, and output the combined contingency information or the combined common information based on the contingency judgment result;

[0082] The label analysis module is configured with a combination analysis strategy, which includes: combining the remaining label information in pairs so that each combination exists, thereby obtaining multiple label combinations; it should be noted that, taking the four types of label information as 1, 2, 3, and 4 as an example, making each combination exist means that a total of 6 combinations (1, 2), (1, 3), (1, 4), (2, 3), (2, 4), and (3, 4) all exist;

[0083] Use the large language model to analyze the two tag information in the tag combination to determine whether the two tag information are contained in a relationship or point to the same meaning. It should be noted that, for example, there are two different tag information "sports" and "basketball", where the sports tag contains the basketball tag. At this time, the tag combination (sports, basketball) has no analytical significance and needs to be deleted.

[0084] Take two different tag information as an example: the two tags "game competition" and "e-sports event" refer to the same competition, that is, the same meaning, and this tag combination also needs to be deleted;

[0085] When the judgment result is yes, the label combination is deleted; when the judgment result is no, no processing is performed.

[0086] The tag analysis module is also configured with a reasonable analysis strategy, which includes: marking the two tag information in the tag combination as the first tag and the second tag respectively; querying the historical viewing information with the first tag as the index; marking all the historical viewing information obtained by the query with 1; querying the historical viewing information with the second tag as the index; marking all the historical viewing information obtained by the query with 2; analyzing the historical viewing information, if there is no historical viewing information marked with 1 and 2 at the same time, outputting invalid combination information; if there is, marking the historical viewing information as combined viewing information, and outputting valid combination information.

[0087] The tag analysis module is also configured with an accidental analysis strategy, which includes: making an accidental judgment on the tag combination of combined valid information, and the accidental judgment includes: marking the largest occurrence frequency in the tag combination as the main frequency, counting the total number of combined viewing information corresponding to the tag combination, and marking it as the analysis frequency; calculating the ratio of the analysis frequency to the main frequency, and marking it as the frequency ratio; using the first tag and the second tag in the tag combination as indexes for querying at the same time, calculating the difference in upload time of two adjacent query results, and marking it as the time interval;

[0088] The accidental analysis formula is used to calculate the frequency ratio and time interval, and the accidental judgment value is output; the accidental analysis formula is configured as: ; Wherein Ro is the accidental judgment value, Pz is the frequency ratio value, Tz is the time interval, and k is a constant; the accidental analysis value is analyzed, and if the accidental analysis value is greater than or equal to the accidental threshold, the combined common information is output; if the accidental analysis value is less than the accidental threshold, the combined accidental information is output; in the specific implementation, k is set to 24, and the accidental threshold is set to 10. Considering the extreme case, there are 10 videos in the second amount of historical viewing information of the user that contain two label information in the label combination at the same time. At this time, it can be judged that the video containing two label information at the same time is not accidental, and the user is more interested in the video containing two label information at the same time;

[0089] In another extreme case, there is only one video in the second amount of historical viewing information that contains both types of label information in a label combination. In this case, k is set to 24, which means that the analysis is performed in units of one day. If a video containing both types of label information is uploaded every 2.9 hours in a day, it can also be proved that the videos containing both types of label information are not accidental.

[0090] The video push module is used to recommend videos to users based on the tag combination of common information;

[0091] The video push module is configured with a video push strategy, which includes: marking the tag combination of common information as a priority combination; analyzing the video information stored in the cloud in real time, and when the stored video information includes two tag combinations in the priority combination at the same time, marking the video information as a priority video; and pushing priority videos to users first.

[0092] Example 2, please refer to Figure 2 As shown, the present invention also provides a video data recommendation method based on tag analysis, comprising:

[0093] Step S1, obtaining a first amount of historical viewing information, the historical viewing information including duration information and tag information; calculating the duration information and outputting the viewing rate; calculating and analyzing the viewing rate, and outputting the effective rate and the effective ratio based on the calculation and analysis results; Step S1 also includes the following sub-steps:

[0094] Step S101, obtaining a first amount of historical viewing information, calculating the ratio of the viewing time of each historical viewing information to the video time, and marking it as a viewing rate;

[0095] Step S102, executing an effective calculation cycle, and outputting an effective rate and an effective ratio based on the effective calculation cycle;

[0096] Step S102 also includes the following sub-steps:

[0097] Step S1021, setting the reference rate with an initial value of 100% and the effective number with an initial value of 0;

[0098] Step S1022, comparing each viewing rate with the reference rate. When the viewing rate is less than the reference rate, no processing is performed; when the viewing rate is greater than or equal to the reference rate, the effective number is increased by one.

[0099] Step S1023, calculating the ratio of the effective number to the first number, marked as analysis rate;

[0100] Step S1024, when the analysis rate is greater than or equal to the first frequency, the reference rate is marked as effective, the analysis rate is marked as effective proportion, and the effective calculation cycle ends; when the analysis rate is less than the first frequency, the effective number is reset to 0, the value obtained by subtracting the second frequency from the reference rate is assigned to the reference rate, and the effective calculation cycle is executed again.

[0101] Step S2, reacquiring historical viewing information based on the effective ratio, and filtering and processing the historical viewing information based on the effective ratio; S2 also includes the following sub-steps:

[0102] Step S201, calculating the product of the first quantity and the effective proportion, and marking it as the first effective number;

[0103] Step S202, calculating the difference between the second number and the first effective number, and marking it as the required number;

[0104] Step S203, calculating the quotient of the required number and the effective proportion, and marking it as the number to be obtained again;

[0105] Step S204, obtaining the historical viewing information of the number of re-acquired items, and calculating the viewing rate of each historical viewing information;

[0106] Step S205: delete the historical viewing information whose viewing rate is less than the effective rate.

[0107] Step S3, statistics are performed on each type of tag information, the statistical results are calculated, and the tag proportion is output; the tag proportion is calculated and analyzed, the low-frequency proportion is output, and the statistical results are screened based on the low-frequency proportion; S3 also includes the following sub-steps:

[0108] Step S301, obtaining the number of remaining historical viewing information, marked as the number of samples;

[0109] Step S302, counting each tag information appearing in all historical viewing information, and marking the counting result as the occurrence frequency;

[0110] Step S303, calculating the ratio of each occurrence frequency to the number of samples, and marking it as label ratio;

[0111] Step S304: mark the first zoom multiple of the largest tag proportion as the low-frequency proportion, and delete the tag information corresponding to the tag proportion that is smaller than the low-frequency proportion.

[0112] Step S4, combining the filtered label information to obtain multiple label combinations; using the large language model to analyze the label combinations, deleting or not processing the label combinations based on the analysis results; querying historical viewing information based on the label combinations, analyzing the query results, and outputting combination valid information and combination invalid information based on the analysis results; S4 also includes the following sub-steps:

[0113] Step S401, combining the remaining label information in pairs so that each combination exists, and obtaining multiple label combinations;

[0114] Step S402, using a large language model to analyze two tag information in the tag combination to determine whether the two tag information have an inclusion relationship or point to the same meaning;

[0115] Step S403, when the judgment result is yes, the tag combination is deleted; when the judgment result is no, no processing is performed;

[0116] Step S404: mark the two tag information in the tag combination as a first tag and a second tag respectively; query the historical viewing information with the first tag as an index, and mark all the historical viewing information obtained by the query as 1;

[0117] Step S405, querying the historical viewing information with the second tag as the index, and marking all the historical viewing information obtained by the query with 2;

[0118] Step S406, analyzing the historical viewing information, if there is no historical viewing information with both 1 and 2, outputting combination invalid information; if there is, marking the historical viewing information as combined viewing information, and outputting combination valid information.

[0119] Step S5, performing a contingency judgment on the tag combination of the combined effective information, and outputting the combined contingency information or the combined common information based on the contingency judgment result; and giving priority to recommending videos to the user based on the tag combination of the combined common information; S5 also includes the following sub-steps:

[0120] Step S501, performing a contingency judgment on the tag combination of combined valid information, the contingency judgment includes:

[0121] Step S502, marking the largest occurrence frequency in the tag combination as the main frequency, counting the total number of combined viewing information corresponding to the tag combination, and marking it as the analysis frequency;

[0122] Step S503, calculating the ratio of the analysis frequency to the main frequency, marked as the frequency ratio;

[0123] Step S504, using the first tag and the second tag in the tag combination as indexes to perform a query, and calculating the difference between the upload times of two adjacent query results, and marking it as a time interval;

[0124] Step S505, using the accidental analysis formula to calculate the frequency ratio and the time interval, and outputting the accidental judgment value;

[0125] The accidental analysis formula is configured as: ; Where Ro is the accidental judgment value, Pz is the frequency ratio value, Tz is the time interval, and k is a constant;

[0126] Step S506, analyzing the accidental analysis value, if the accidental analysis value is greater than or equal to the accidental threshold, outputting the combined common information; if the accidental analysis value is less than the accidental threshold, outputting the combined accidental information;

[0127] Step S507, marking the tag combination combining common information as a priority combination;

[0128] Step S508, analyzing the video information stored in the cloud in real time, when the stored video information includes two tag combinations in the priority combination at the same time, marking the video information as a priority video; and pushing the priority video to the user first.

[0129] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0130] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0131] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

Claims

1. A video data recommendation method based on label analysis, characterized in that: include: Acquire a first amount of historical viewing information, where the historical viewing information includes duration information and tag information; Calculate the duration information and output the viewing rate; Calculate and analyze the viewing rate, and output the effective rate and effective ratio based on the calculation and analysis results; Based on the effective ratio, historical viewing information is obtained for a second time, and the historical viewing information is filtered and processed based on the effective ratio; Perform statistics on each type of tag information, calculate the statistical results, and output the tag ratio; perform calculation and analysis on the tag ratio, output the low-frequency ratio, and filter the statistical results based on the low-frequency ratio; The filtered label information is combined to obtain multiple label combinations; the label combinations are analyzed using a large language model, and the label combinations are deleted or not processed based on the analysis results; Query historical viewing information based on the tag combination, analyze the query results, and output combination valid information and combination invalid information based on the analysis results; Perform accidental judgment on the combination of labels of combined effective information, and output combined accidental information or combined common information based on the accidental judgment result; and give priority to recommending videos to users based on the combination of labels of combined common information; The duration information includes video duration and viewing duration; Obtaining a first amount of historical viewing information, calculating the duration information, and outputting a viewing rate; Calculate and analyze the viewing rate, and output the efficiency and effective ratio based on the calculation and analysis results, including: Obtain a first amount of historical viewing information, calculate a ratio of viewing time of each piece of historical viewing information to video time, and mark the ratio as viewing rate; Execute an effective calculation cycle, and output the effective rate and effective ratio based on the effective calculation cycle; The effective calculation cycle includes: Set the reference rate with an initial value of 100% and the effective number with an initial value of 0; Compare each viewing rate with the reference rate. When the viewing rate is less than the reference rate, no processing is done; when the viewing rate is greater than or equal to the reference rate, the effective number is increased by one. Calculate the ratio of the significant number to the first number, labeled as the analysis ratio; When the analysis rate is greater than or equal to the first frequency, the reference rate is marked as effective, the analysis rate is marked as effective proportion, and the effective calculation cycle ends; When the analysis rate is less than the first frequency, the effective number is reset to 0, the value obtained by subtracting the second frequency from the reference rate is assigned to the reference rate, and the effective calculation cycle is performed again; Statistics are collected for each type of tag information, the statistical results are calculated, and the tag ratio is output; the tag ratio is calculated and analyzed, the low-frequency ratio is output, and the statistical results are filtered based on the low-frequency ratio, including: Get the number of remaining historical viewing information, marked as the number of samples; Count each tag information that appears in all historical viewing information, and mark the count result as the occurrence frequency; Calculate the ratio of each occurrence frequency to the number of samples, and mark it as the label ratio; The first scaling factor of the largest label ratio is marked as the low-frequency ratio, and the label information corresponding to the label ratio smaller than the low-frequency ratio is deleted; Perform accidental judgment on the label combination of combined valid information, and output combined accidental information or combined common information based on the accidental judgment result, including: The combination of tags that combine valid information is subjected to a contingency judgment, wherein the contingency judgment includes: Mark the largest occurrence frequency in the tag combination as the main frequency, count the total number of combined viewing information corresponding to the tag combination, and mark it as the analysis frequency; Calculate the ratio of the analysis frequency to the main frequency, marked as the frequency ratio; The first tag and the second tag in the tag combination are used as indexes for querying, and the difference between the upload times of two adjacent query results is calculated and marked as the time interval; Use the accidental analysis formula to calculate the frequency ratio and time interval, and output the accidental judgment value; The accidental analysis formula is configured as: ; Where Ro is the accidental judgment value, Pz is the frequency ratio value, Tz is the time interval, and k is a constant; The accidental analysis value is analyzed. If the accidental analysis value is greater than or equal to the accidental threshold, the combined common information is output; if the accidental analysis value is less than the accidental threshold, the combined accidental information is output.

2. The video data recommendation method based on tag analysis according to claim 1, characterized in that: Based on the effective ratio, historical viewing information is obtained for the second time, and the historical viewing information is filtered and processed based on the efficiency, including: Calculate the product of the first quantity and the effective proportion, and mark it as the first effective number; Calculate the difference between the second number and the first effective number, marking it as the required number; Calculate the quotient of the required number and the effective proportion, and mark it as the number to be obtained again; Obtain the historical viewing information of the number of re-obtained items, and calculate the viewing rate of each historical viewing information; The historical viewing information with a viewing rate lower than the effective rate will be deleted.

3. The video data recommendation method based on label analysis according to claim 2, characterized in that: The filtered label information is combined to obtain multiple label combinations; the label combinations are analyzed using a large language model, and the label combinations are deleted or not processed based on the analysis results, including: Combine the remaining label information in pairs so that each combination exists, and obtain multiple label combinations; Use the large language model to analyze the two tag information in the tag combination to determine whether the two tag information are included in the relationship or point to the same meaning; When the judgment result is yes, the label combination is deleted; when the judgment result is no, no processing is performed.

4. The video data recommendation method based on label analysis according to claim 3 is characterized in that: The historical viewing information is queried based on the tag combination, the query result is analyzed, and the combination valid information and combination invalid information are output based on the analysis result, including: Marking two tag information in the tag combination as a first tag and a second tag respectively; querying the historical viewing information using the first tag as an index; All the historical viewing information obtained by query is marked as 1; Query the historical viewing information using the second tag as an index; All the historical viewing information obtained by query is marked as 2; The historical viewing information is analyzed. If there is no historical viewing information marked with both 1 and 2, the combination invalid information is output; if there is, the historical viewing information is marked as combined viewing information, and the combination valid information is output.

5. The video data recommendation method based on label analysis according to claim 4, characterized in that: The video recommendations for users are based on the combination of tags with common information, including: Mark the tag combinations that combine common information as priority combinations; Analyze the video information stored in the cloud in real time, and when the stored video information includes two tag combinations in the priority combination at the same time, mark the video information as a priority video; Push priority videos to users first.

6. A video data recommendation system based on tag analysis, used to implement a video data recommendation method based on tag analysis according to any one of claims 1 to 5, characterized in that: Including habit analysis module, tag statistics module, tag analysis module and video push module; The habit analysis module is used to obtain a first amount of historical viewing information, wherein the historical viewing information includes duration information and tag information; calculate the duration information and output the viewing rate; calculate and analyze the viewing rate, and output the effective rate and effective ratio based on the calculation and analysis results; The habit analysis module is also used to obtain historical viewing information based on the effective ratio and filter the historical viewing information based on the effective rate; The tag statistics module is used to count each tag information, calculate the statistical results, and output the tag proportion; calculate and analyze the tag proportion, output the low-frequency proportion, and filter the statistical results based on the low-frequency proportion; The tag analysis module is used to combine the tag information after screening to obtain multiple tag combinations; use the large language model to analyze the tag combinations, and delete or do not process the tag combinations based on the analysis results; Query historical viewing information based on the tag combination, analyze the query results, and output combination valid information and combination invalid information based on the analysis results; The label analysis module is also used to make a contingency judgment on the label combination of the combined valid information, and output the combined contingency information or the combined common information based on the contingency judgment result; The video push module is used to preferentially recommend videos to users based on tag combinations that combine common information.

Images