A quality detection system for OTT based on human-computer interaction

Abstract

The present invention discloses an OTT-oriented quality inspection system based on human-computer interaction, and relates to the technical field of multimedia quality inspection. The system comprises: a facial expression capture module, an emotion analysis module, a feedback module, a data processing module and a problem processing module. The present invention can accurately capture and analyze the facial expressions of users when watching OTT content through advanced algorithm formulas, so as to perceive the emotional state of users in real time. This feature enables the system to respond quickly when users show negative emotions, and guide users to feedback quality problems by actively popping up an inquiry window and using friendly prompts, which not only improves the timeliness of quality inspection, but also ensures the accuracy and completeness of user feedback. This intelligent interaction method allows users to experience a more intimate and efficient service experience, and also provides valuable user feedback data for OTT content providers, which helps them continuously optimize content quality.

Description

Technical Field

[0001] The present invention relates to the technical field of multimedia quality detection, and in particular to an OTT-oriented quality detection system based on human-computer interaction. Background Art

[0002] With the rapid development and popularization of the Internet, OTT (Over The Top) service, as a service model for transmitting video and audio content through the Internet, has gradually become an indispensable part of people's daily life. The OTT platform not only provides users with rich film and television resources, but also improves users' viewing experience through continuous technological innovation. However, with the continuous expansion of the user scale and the increasing variety of content, how to ensure the quality of OTT content and promptly discover and solve problems encountered by users during the viewing process has become an important issue that needs to be urgently addressed in the current OTT service field.

[0003] In traditional OTT quality detection, although the interactive feedback mechanism of emotion perception has been applied to a certain extent, it still has many shortcomings. Although the emotion analysis algorithm in artificial intelligence technology can be used to capture the facial expressions of users when watching OTT content through the camera of smart devices, thereby analyzing the user's emotional state, this mechanism can detect the user's negative emotions and actively pop up an inquiry window to ask the user whether he has encountered quality problems, and provide corresponding feedback options for the user to choose. Such a mechanism has achieved real-time monitoring of the user's viewing experience to a certain extent, and can provide timely feedback when problems are found.

[0004] However, the existing emotion-aware interactive feedback mechanism still has some significant problems. First, the accuracy of facial expression capture and emotion analysis needs to be improved. Due to individual differences in facial expressions and reaction patterns of different users, how to accurately capture and analyze these differences is a major challenge facing current technology. Secondly, traditional emotion analysis algorithms often lack sufficient consideration of user viewing history and preferences, resulting in the inability to accurately judge the user's true emotional state in some cases. In addition, traditional mechanisms also have certain limitations in the real-time and effectiveness of emotional feedback. They are unable to capture all emotional changes of users in a timely manner, nor can they be flexibly adjusted and improved based on user feedback.

[0005] Therefore, it is necessary to develop an OTT-oriented quality inspection system based on human-computer interaction to provide more reliable and efficient quality inspection services for the OTT industry and promote the sustainable and healthy development of the industry. Summary of the invention

[0006] The purpose of the present invention is to make up for the shortcomings of the prior art and to provide an OTT-oriented quality inspection system based on human-computer interaction. It can capture and analyze the facial expressions of users when watching OTT content in real time through an intelligent camera using an algorithm formula to judge the user's emotional state. When negative emotions are detected, the system will actively pop up an inquiry window and provide friendly feedback options, thereby improving user participation and feedback accuracy. The system can also collect, organize and analyze user feedback, predict the development trend of quality problems, and take corresponding solutions.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a quality detection system for OTT based on human-computer interaction, the system comprising: a facial expression capture module, an emotion analysis module, a feedback module, a data processing module and a problem processing module;

[0008] The facial expression capture module: relies on a smart camera to capture the user's face in real time when watching OTT content. The camera has high resolution and frame rate for capturing subtle changes, and has the function of automatically adjusting parameters and is well compatible with the operating system;

[0009] The emotion analysis module: uses an algorithm formula to conduct in-depth analysis on the facial image data obtained by the facial expression capture module. The algorithm formula can identify multiple emotional states after being trained with a large amount of data. It can judge each emotional state through the emotional feature value and has the ability to quickly process a large amount of data;

[0010] The emotion analysis module quantifies the emotion features in the facial image through the emotion feature quantization formula. Assuming the coordinates of a feature point in the facial image are (x, y), its brightness value is I(x, y), and the calculation formula for defining the emotion feature quantization value Q is: Where n and m are the number of pixels in the width and height directions of the image, respectively, and w ij is a weight coefficient pre-set according to different emotional states;

[0011] The emotion analysis module is used to judge the emotion state corresponding to the facial expression through the emotion state judgment formula, and the emotion state judgment formula is: S=αP+βF+γT, where P is the positive emotion feature quantization value of the facial expression, F is the negative emotion feature quantization value, T is the neutral emotion feature quantization value, α, β, γ are the corresponding weight coefficients, and α+β+γ=1, and the emotion state is judged according to the score. When S is greater than a certain threshold, it is judged as a positive emotion state, when it is less than another threshold, it is judged as a negative emotion state, and when it is between the two thresholds, it is judged as a neutral emotion state;

[0012] The feedback module: when the sentiment analysis module detects negative sentiments of the user, it is immediately activated, and a query window is actively popped up on the screen of the smart device to ask the user in a friendly way whether he encounters quality problems, and provides feedback options for video freezes, unclear picture quality, abnormal sound, and subtitle problems. The feedback options are concise and easy for users to understand and choose, and the user participation and feedback accuracy are improved by setting a reward mechanism;

[0013] The data processing module is responsible for collecting, organizing and analyzing user feedback information, classifying and counting user feedback issues to determine the type and severity of quality issues, and combining historical data for analysis to take preventive measures in advance for frequently occurring quality issues;

[0014] The problem handling module: takes corresponding solutions according to the type and severity of the quality problem determined by the data processing module, and also has the ability to automatically detect and repair some common problems, and can adjust and improve the solutions based on user feedback and actual results.

[0015] Furthermore, the emotional state threshold determination in the emotional analysis module is provided with the emotional state scores of n samples, and the average of the emotional state scores is calculated, and the formula is: Calculate the standard deviation of the affective state scores using the formula: Set the positive emotional state threshold T pos =S avg +k1σ, negative affective state threshold T neg =S avg -k2σ, where k1 and k2 are adjustment coefficients, and k1 and k2 are greater than 0. When S ≥ T pos When S≤T neg When T neg <S<T pos , it is judged as a neutral emotional state.

[0016] Furthermore, the facial feature points related to the positive, negative and neutral emotional feature quantities in the sentiment analysis module are:

[0017] Facial feature points related to positive emotional features: curved eyes, upturned corners of mouth, relaxed facial muscles, and raised apple cheeks;

[0018] Negative emotional feature related facial feature points: frowning, downturned corners of mouth, angry eyes, biting lips;

[0019] Facial feature points related to neutral emotional features: calm facial expression, peaceful eyes, and naturally closed lips.

[0020] Furthermore, when analyzing the facial expression data, the emotion analysis module also considers the user's viewing history and preference factors. The emotion analysis module records the historical emotion state data of the user when viewing different types of OTT content. Let the number of times the user views a certain specific type of content be N type , where the number of times the positive emotion state appears is N pos , the number of times the negative emotion state appears is N neg , and the number of times the neutral emotion state appears is N neu . Define the positive emotion tendency degree under this type of content as: The negative emotion tendency degree is: The neutral emotion tendency degree is: When analyzing the current facial expression data, adjust the emotion state judgment by combining these tendency degrees. When the user has a high positive emotion tendency degree for the current content type being viewed, appropriately increase the weight of the positive emotion feature quantization value P, that is, the adjusted emotion state score becomes: S ′ =α ′ P + β ′ F + γ ′ T, where α ′ >α, β ′ ≤β, γ ′ ≤γ. Similarly, make corresponding adjustments for the cases of high negative emotion tendency degree and neutral emotion tendency degree.

[0021] Furthermore, the emotion analysis module predicts the development direction of the user's emotion state by establishing an emotion state change model. Let the emotion state scores at consecutive time points be S1, S2, S3, …, S n . Define the emotion state change speed as: where t1, t2, …, t n are the corresponding time points, and the emotion change amplitude is: A = |S n - S1|. When it is detected that the user's emotion state gradually changes from happy to dissatisfied, that is, S n < S1 and meets a certain change amplitude threshold A threshold , give an early warning of possible quality problems and promptly notify the problem handling module to take corresponding measures.

[0022] Furthermore, the emotion analysis module adjusts the parameters of the algorithm formula according to different OTT content types. For different types of content such as movies, TV dramas, and variety shows, there are differences in the user's emotional reactions. Let different OTT content types be T1 movies, T2 TV dramas, and T3 variety shows respectively. For each content type, define an emotion adjustment coefficient vector Among them, C1 corresponds to the adjustment coefficient of positive emotional characteristics, C2 corresponds to the adjustment coefficient of negative emotional characteristics, and C3 corresponds to the adjustment coefficient of neutral emotional characteristics. For the emotional state score formula: S = αP + βF + γT, when analyzing different content types, it is adjusted to: S T =α×C 1T P+β×C 2T F+γ×C 3T T, where S T represents the sentiment state score for a specific content type T.

[0023] Furthermore, the sentiment analysis module also calculates the user's emotional state through the sentiment intensity calculation formula to measure the degree to which the user's emotion deviates from the neutral state. Let the sentiment intensity be E, the positive sentiment feature quantization value be P, the negative sentiment feature quantization value be F, and the neutral sentiment feature quantization value be T. The sentiment intensity calculation formula is defined as:

[0024] Furthermore, the data processing module determines the priority of the quality problem through a problem severity calculation formula. Assuming the number of feedbacks of a certain quality problem is N, the number of affected users is U, and the problem severity S calculation formula is defined as: Where N total is the total number of feedbacks, U total is the total number of users, k is the adjustment coefficient, and the feedback issues are sorted according to the severity S value of the issue.

[0025] Furthermore, the data processing module predicts the development trend of quality problems through the problem trend prediction formula. Suppose the problem trend prediction value is T, and the problem severity at the current time point is S. t The severity of the problem in the previous period is S t-1 , the problem trend prediction formula is: Where t is the current time, t last is the end time of the previous time period, k is the adjustment coefficient, and a positive problem trend prediction value indicates that the problem may be worsening, while a negative problem trend prediction value indicates that the problem has improved.

[0026] Furthermore, the problem processing module evaluates the effect of problem solving through the problem solving effect evaluation formula, assuming that the problem solving effect evaluation value is E, and the user's negative emotion intensity before the problem is solved is E before , the user's negative emotion intensity after the problem is solved is E after , the problem solving effect evaluation formula is: The higher the problem-solving effect evaluation value, the better the problem-solving effect.

[0027] Compared with the existing technology, this OTT-oriented quality detection system based on human-computer interaction has the following beneficial effects:

[0028] 1. The present invention uses advanced algorithm formulas to accurately capture and analyze the facial expressions of users when watching OTT content, thereby perceiving the emotional state of users in real time. This feature enables the system to respond quickly when users express negative emotions, by actively popping up an inquiry window and guiding users to provide feedback on quality issues with friendly prompts. This not only improves the timeliness of quality inspection, but also ensures the accuracy and completeness of user feedback. This intelligent interactive method allows users to experience a more intimate and efficient service experience, while also providing OTT content providers with valuable user feedback data, which helps them continuously optimize content quality.

[0029] 2. By setting up a reward mechanism, the present invention further stimulates the enthusiasm of users to participate in feedback, thereby ensuring the diversity and representativeness of feedback data. This interactive feedback mechanism based on emotional perception not only helps to timely discover and solve quality problems in OTT content, but also can provide strong support for the continuous optimization of system algorithms through continuous accumulation of user feedback data. Ultimately, this will bring users a smoother and higher-quality viewing experience, and also contribute to the healthy development of the OTT industry.

[0030] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 It is a flow chart of a quality inspection system for OTT based on human-computer interaction;

[0033] Figure 2 Flowchart of the sentiment analysis module. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. 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.

[0035] Embodiment 1:

[0036] User experience optimization of video playback platform:

[0037] A large video playback platform hopes to improve user experience, especially to promptly understand user feedback on video quality and be able to quickly resolve problems. The platform has a large number of movie, TV series, and variety show resources, and a wide user base covering different age groups and regions.

[0038] Facial expression capture module. When the user watches a video, the camera on the smart device (such as a mobile phone, tablet computer, smart TV) starts the facial expression capture module. The camera automatically adjusts the sensitivity, contrast and brightness parameters according to the ambient light intensity. At the same time, the camera's resolution and frame rate meet the requirements of capturing subtle facial changes, such as being able to clearly capture the slight upward movement of the user's mouth corners and frowning expressions.

[0039] The emotion analysis module receives the facial image data obtained by the facial expression capture module. First, the emotion feature quantification formula is used: Quantify the image data. For example, for a comedy movie, when the user's eyes are curved and the corners of his mouth are raised, the weight coefficient w of the corresponding feature point is ij So that the positive emotional feature quantization value P increases, let the coordinates of a feature point in the facial image be (x i ,y j ), whose brightness value is I(x i ,y j ), n and m are the number of pixels in the width and height directions of the image respectively. Then, according to the emotional state judgment formula: S = αP + βF + γT, the emotional state is judged. For this comedy movie, the positive emotional state threshold T is set according to historical data. pos =S avg +k1σ, negative affective state threshold T neg =S avg -k2σ(where is the average of the emotional state scores, is the standard deviation of the emotional state score, k1 and k2 are adjustment coefficients), if the calculated emotional state score S is greater than T pos, the user is judged to be in a positive emotional state, indicating that the user is satisfied with the video content. The user's viewing history and preference factors will also be considered. If the user often watches comedy movies and the historical emotional state data shows a positive emotional tendency Higher (assuming the number of times the user watches comedy movies is N type , where the number of times the positive emotional state appears is N pos ), then when analyzing the facial expression data of the current comedy movie, the weight of the positive emotional feature quantization value P is appropriately increased, that is, the adjusted emotional state score becomes: S ′ =α ′ P+β ′ F+γ ′ T, where α ′ >α,γ ′ ≤γ, further accurately judge the user’s emotional state.

[0040] When the sentiment analysis module detects negative emotions from the user, the feedback module pops up an inquiry window on the screen of the smart device. For example, when the user is watching a TV series with unclear picture quality, the feedback module pops up a window to ask the user whether he or she has encountered picture quality problems, and provides feedback options for video freezes, unclear picture quality, abnormal sound, and subtitle problems. The window design is simple and clear, and a reward mechanism, such as point rewards, is set up to encourage users to accurately feedback problems.

[0041] The data processing module collects user feedback information. For feedback on unclear image quality, the severity of the problem is calculated using the formula: Determine the severity of the problem. Assume that the number of feedbacks about the unclear picture quality of the TV series is large, the number of affected users is also large, and the total number of feedbacks is N. t otal and the total number of users U total It is known that the calculated problem severity S is high, and the data processing module will list this problem as a priority. At the same time, the data processing module can also use the problem trend prediction formula as follows: Predict the development trend of the problem. If the number of feedbacks about unclear image quality continues to increase over time, S t (the severity of the problem at the current time point) is greater than S t-1 (the severity of the problem in the previous time period), the problem trend prediction value T is positive, indicating that the problem is worsening and timely measures need to be taken.

[0042] The problem handling module takes corresponding measures according to the severity and type of the problem determined by the data processing module. For the problem of unclear image quality, the resolution, color saturation, and contrast parameters of the video may be adjusted. After the adjustment, the problem solving effect evaluation formula is: To evaluate the effect of problem solving, let the user's negative emotion intensity E before the problem is solved before Higher, the user's negative emotion intensity E after the problem is solved after The calculated problem-solving effect evaluation value E is higher, indicating that the problem-solving measures are effective.

[0043] To summarize, this embodiment can obtain clear facial image data under different lighting environments through automatic parameter adjustment of the facial expression capture module. The emotion analysis module comprehensively uses a variety of formulas and algorithms to accurately judge the user's emotional state, and further improves the accuracy of judgment by considering the user's viewing history and preference factors. The feedback module obtains user feedback in a timely manner, the data processing module reasonably determines the severity and trend of the problem, the problem handling module effectively solves the problem and evaluates the effect. The entire system works in coordination to effectively improve the user experience of the video playback platform and promptly discover and solve video quality problems.

[0044] Embodiment 2:

[0045] Teaching quality monitoring of online education platforms:

[0046] An online education platform provides a variety of courses, including academic courses and skill training courses. The platform hopes to understand the emotional state of students during the learning process so as to adjust the teaching content and methods in a timely manner and improve the teaching quality.

[0047] Facial expression capture module. When students watch online course videos, the camera on the smart device starts the facial expression capture module. The camera has an automatic focus adjustment function. When the distance between the student and the device changes, the focus can be automatically adjusted to ensure clear capture of facial expressions. At the same time, in order to ensure normal operation in different lighting environments, the camera's sensitivity, contrast and brightness parameters can also be automatically adjusted.

[0048] The sentiment analysis module receives the facial image data obtained by the facial expression capture module and performs analysis. For different types of courses, the sentiment analysis module adjusts the parameters of the sentiment analysis algorithm according to the course type. For example, when analyzing the facial expression data of a mathematics course, it focuses on whether the students have confused emotional characteristics. By defining different emotion adjustment coefficient vectors For the Mathematics course (set to T1), set make (Adjustment coefficient corresponding to negative emotional characteristics) is relatively large, because confusion belongs to negative emotions and may be more concerned when learning mathematics. Then, the quantitative formula through emotional characteristics is: Quantify the image data, and assume that the coordinates of a feature point in the facial image are (x i ,y j ), whose brightness value is I(x i ,yj ), where n and m are the number of pixels in the width and height directions of the image respectively. Then, according to the emotional state judgment formula: S = αP + βF + γT, the emotional state of the student is judged. The emotional analysis module will also conduct real-time monitoring and trend analysis on continuous facial expression changes. Let the emotional state scores of the student at consecutive time points be S1, S2, S3, …, S n , the speed of emotional state change (where t1, t2, …, t n are the corresponding time points), the emotional change amplitude A = |S n - S1|. When it is detected that the emotional state of the student gradually changes from understanding to confusion, that is, S n < S1 and meets a certain change amplitude threshold A threshold , an early warning of possible teaching quality problems is given in advance, and the problem handling module is notified in time to take corresponding measures.

[0049] Feedback module. When the emotional analysis module detects negative emotions of the student, the feedback module pops up an inquiry window on the screen of the intelligent device. The window asks the student whether they encounter problems in understanding the teaching content or video playback, and provides corresponding feedback options. At the same time, in order to facilitate the student's feedback, the feedback module can also provide a voice prompt function, especially for students who are inconvenient to view the screen.

[0050] Data processing module. It collects the student feedback information and conducts classification and statistics. For the feedback problems of not understanding the teaching content, the formula for calculating the problem severity is: to determine the severity of the problem. Suppose the number of feedbacks N for not understanding this teaching content is relatively large, and the number of affected students U is also large. The total number of feedbacks N t otal and the total number of students U total are known. If the calculated problem severity S is relatively high, the data processing module will list this problem as a priority. At the same time, the data processing module also uses the formula for predicting the problem trend: to predict the development trend of the problem. If, over time, the number of feedbacks on the problem of not understanding the teaching content continues to increase, S t (the problem severity at the current time point) is greater than S t-1 (the problem severity of the previous multiple segment), and the problem trend prediction value T is positive, indicating that the problem is deteriorating and corresponding measures need to be taken in time.

[0051] Problem handling module. It takes corresponding measures according to the problem severity and type determined by the data processing module. For the problem of not understanding the teaching content, the teaching method may be adjusted, such as increasing case explanations and slowing down the explanation speed. After the adjustment, the formula for evaluating the problem solving effect is: to evaluate the problem solving effect. Suppose the negative emotion intensity of the student before the problem is solved is the solving effect Ebefore Higher, negative emotion intensity E of students after problem solving after The calculated problem-solving effect evaluation value E is higher, which shows that the adjustment of teaching methods is effective and the teaching quality is improved.

[0052] To summarize, this embodiment ensures clear capture of facial images under different lighting and distance conditions through the facial expression capture module. The emotion analysis module accurately judges the emotional state of students by targeted adjustment of algorithm parameters and monitoring and trend analysis of continuous expression changes. The feedback module facilitates students to feedback problems. The data processing module rationally determines the severity and trend of the problem. The problem handling module effectively solves problems and evaluates the effects. The entire system works in coordination, which helps the online education platform to timely understand the students' learning status, improve the teaching quality, and optimize the teaching content and methods.

[0053] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A quality detection system for OTT based on human-computer interaction, characterized in that: The system includes: a facial expression capture module, an emotion analysis module, a feedback module, a data processing module and a problem processing module; The facial expression capture module: relies on a smart camera to capture the user's face in real time when watching OTT content. The camera has high resolution and frame rate for capturing subtle changes, and has the function of automatically adjusting parameters and is well compatible with the operating system; The emotion analysis module: uses an algorithm formula to conduct in-depth analysis on the facial image data obtained by the facial expression capture module. The algorithm formula can identify multiple emotional states after being trained with a large amount of data. It can judge each emotional state through the emotional feature value and has the ability to quickly process a large amount of data; The emotion analysis module quantifies the emotion features in the facial image through the emotion feature quantization formula. Assuming the coordinates of a feature point in the facial image are (x, y), its brightness value is I(x, y), and the calculation formula for defining the emotion feature quantization value Q is: Where n and m are the number of pixels in the width and height directions of the image, respectively, and w ij is a weight coefficient pre-set according to different emotional states; The emotion analysis module is used to judge the emotion state corresponding to the facial expression through the emotion state judgment formula, and the emotion state judgment formula is: S=αP+βF+γT, where P is the positive emotion feature quantization value of the facial expression, F is the negative emotion feature quantization value, T is the neutral emotion feature quantization value, α, β, γ are the corresponding weight coefficients, and α+β+γ=1, and the emotion state is judged according to the score. When S is greater than a certain threshold, it is judged as a positive emotion state, when it is less than another threshold, it is judged as a negative emotion state, and when it is between the two thresholds, it is judged as a neutral emotion state; The feedback module: when the sentiment analysis module detects negative sentiments of the user, it is immediately activated, and a query window is actively popped up on the screen of the smart device to ask the user in a friendly way whether he encounters quality problems, and provides feedback options for video freezes, unclear picture quality, abnormal sound, and subtitle problems. The feedback options are concise and easy for users to understand and choose, and the user participation and feedback accuracy are improved by setting a reward mechanism; The data processing module is responsible for collecting, organizing and analyzing user feedback information, classifying and counting user feedback issues to determine the type and severity of quality issues, and combining historical data for analysis to take preventive measures in advance for frequently occurring quality issues; The problem handling module: takes corresponding solutions according to the type and severity of the quality problem determined by the data processing module, and also has the ability to automatically detect and repair some common problems, and can adjust and improve the solutions based on user feedback and actual results.

2. According to claim 1, a quality detection system for OTT based on human-computer interaction is characterized in that: The emotional state threshold determination in the emotional analysis module is provided with the emotional state scores of n samples, and the average emotional state scores are calculated, and the formula is: Calculate the standard deviation of the affective state scores using the formula: Set the threshold of positive emotional state T pos =S avg +k1σ, negative affective state threshold T neg =S avg -k2σ, where k1 and k2 are adjustment coefficients, and k1 and k2 are greater than 0. When S ≥ T pos When S≤T neg When T neg <S<T pos , it is judged as a neutral emotional state.

3. According to claim 1, a quality detection system for OTT based on human-computer interaction is characterized in that: The facial feature points related to the positive, negative and neutral emotional feature quantities in the sentiment analysis module are: Facial feature points related to positive emotional features: curved eyes, upturned corners of mouth, relaxed facial muscles, and raised apple cheeks; Negative emotional feature related facial feature points: frowning, downturned corners of mouth, angry eyes, biting lips; Facial feature points related to neutral emotional features: calm facial expression, peaceful eyes, and naturally closed lips.

4. According to claim 1, a quality detection system for OTT based on human-computer interaction is characterized in that: The emotion analysis module also considers the user's viewing history and preference factors when analyzing facial expression data. The emotion analysis module records the user's historical emotional state data when watching different types of OTT content. Suppose the number of times a user watches a certain type of content is N. type , where the number of times the positive emotional state appears is N pos , the number of times the negative emotional state occurs is N neg , the number of times the neutral emotional state appears is N neu , define the positive sentiment tendency under this type of content as: Negative sentiment tendency is: Neutral sentiment tendency is: When analyzing the current facial expression data, the emotional state judgment is adjusted in combination with these tendencies. When the user has a high positive emotional tendency towards the type of content currently being watched, the weight of the positive emotional feature quantization value P is appropriately increased, that is, the adjusted emotional state score becomes: S ′ =α ′ P+β ′ F+γ ′ T, where α ′ >α,β ′ ≤β,γ ′ ≤γ. Similarly, corresponding adjustments are made for situations with higher negative and neutral emotional tendencies.

5. According to claim 1, a quality detection system for OTT based on human-computer interaction is characterized in that: The emotional analysis module predicts the development direction of the user's emotional state by establishing an emotional state change model. Let the emotional state scores at consecutive time points be S1, S2, S3, …, S n , and define the emotional state change speed as: where t1, t2, …, t n are the corresponding time points, and the emotional change amplitude is: A = |S n - S1|. When it is detected that the user's emotional state gradually changes from happy to dissatisfied, that is, S n < S1 and satisfies a certain change amplitude threshold A threshold , an early warning of possible quality problems is given, and the problem handling module is notified in a timely manner to take corresponding measures.

6. The quality detection system for OTT based on human-computer interaction according to claim 1 is characterized in that: The sentiment analysis module adjusts the parameters of the algorithm formula according to different OTT content types. For different types of content such as movies, TV series, and variety shows, users have different emotional responses. Assume that different OTT content types are T1 movies, T2 TV series, and T3 variety shows. For each content type, define an emotion adjustment coefficient vector Among them, C1 corresponds to the adjustment coefficient of positive emotional characteristics, C2 corresponds to the adjustment coefficient of negative emotional characteristics, and C3 corresponds to the adjustment coefficient of neutral emotional characteristics. For the emotional state score formula: S = αP + βF + γT, when analyzing different content types, it is adjusted to: Where S T represents the sentiment state score for a specific content type T.

7. The OTT-oriented quality detection system based on human-computer interaction according to claim 1 is characterized in that: The sentiment analysis module also calculates the user's emotional state through the sentiment intensity calculation formula to measure the degree to which the user's emotion deviates from the neutral state. Let the sentiment intensity be E, the positive sentiment feature quantization value be P, the negative sentiment feature quantization value be F, and the neutral sentiment feature quantization value be T. The sentiment intensity calculation formula is defined as:

8. The OTT-oriented quality detection system based on human-computer interaction according to claim 1 is characterized in that: The data processing module determines the priority of the quality problem through the problem severity calculation formula. Assuming the number of feedbacks of a quality problem is N, the number of affected users is U, and the problem severity S calculation formula is defined as: Where N total is the total number of feedbacks, U total is the total number of users, k is the adjustment coefficient, and the feedback issues are sorted according to the severity S value of the issue.

9. The OTT-oriented quality detection system based on human-computer interaction according to claim 1 is characterized in that: The data processing module predicts the development trend of quality problems through the problem trend prediction formula. The problem trend prediction value is T, and the problem severity at the current time point is S. t The severity of the problem in the previous period is S t-1 , the problem trend prediction formula is: Where t is the current time, t last is the end time of the previous time period, k is the adjustment coefficient, and a positive problem trend prediction value indicates that the problem may be worsening, while a negative problem trend prediction value indicates that the problem has improved.

10. The OTT-oriented quality detection system based on human-computer interaction according to claim 1, characterized in that: The problem processing module evaluates the effect of problem solving through the problem solving effect evaluation formula. The problem solving effect evaluation value is set to E, and the user's negative emotion intensity before the problem is solved is E before , the user's negative emotion intensity after the problem is solved is E after , the problem solving effect evaluation formula is: The higher the problem-solving effect evaluation value, the better the problem-solving effect.

Images