Evaluation method of overseas communication effect of film and television works based on multimodal large model

Abstract

The present invention relates to the field of text information processing technology, and in particular to a method for evaluating the overseas dissemination effect of film and television works based on a multimodal large model. The method evaluates film and television works based on existing processed viewing data to obtain an initial evaluation effect. The guiding emphasis of each derived evaluation feature vector in the derivative video is determined by a matching method, and the derivative guiding vector of the derivative video can be screened out. The theme content vector of the derivative video is extracted and matched with the derived guiding vector, and the guiding theme vector and the guiding score are determined according to the degree of matching. The matching degree and guiding score of the same derived theme vector in all derivative videos are counted, and the impression tendency score of each guiding theme vector is determined, the derivative video influence score is obtained, and the initial evaluation effect is adjusted. The present invention quantifies the influence of the theme content in the derivative video through the information matching method, and obtains a final evaluation effect that can accurately characterize the dissemination effect of the film and television work.

Description

Technical Field

[0001] The present invention relates to the technical field of text information processing, and in particular to a method for evaluating the overseas dissemination effect of film and television works based on a multimodal large model. Background Art

[0002] In the context of globalization, film and television, as important carriers of culture, have a direct impact on the shaping and dissemination of cultural imagery through their overseas dissemination. Through the in-depth integration of multimodal narratives (text, images, sound, and visual symbols), film and television not only carry local cultural values ​​but also serve the dual function of cultural decoding and reconstruction in cross-cultural communication.

[0003] The current distribution of film and television works is fragmented and multi-platform, especially when they are exported overseas. Due to cultural differences between regions, the dissemination of film and television works is even more dependent on the fragmented dissemination of information through numerous platforms and short videos. When film and television works are distributed overseas, overseas distributors and theater chains will localize the works according to the local language in the hope that audiences in the region can better enjoy the works. However, due to cultural differences between different regions, the localized works released may not fully conform to the cultural habits of the current region, making it difficult for audiences in the corresponding regions to quickly understand the content of the film and television works. As a result, many derivative videos such as clips, explanations, and text descriptions are generated about the film and television works. By viewing these derivative videos, audiences can understand and digest the true core culture of the film and television works and provide a true evaluation of the works.

[0004] Existing methods for evaluating the communication effectiveness of film and television works, such as the one proposed in Publication No. CN112085390A, use different weights assigned to different evaluation indicators during different release periods to comprehensively calculate the communication effectiveness assessment value. This method essentially evaluates the target film and television work based on existing viewing data and fails to consider the influence of derivative videos, which can guide audience understanding and evaluation. Existing technologies that evaluate the communication effectiveness of a target film and television work based solely on existing viewing data cannot accurately represent the target film and television work's true communication effectiveness. Summary of the Invention

[0005] In order to solve the technical problem that the existing technology only analyzes viewing data during the evaluation of target film and television works, but does not analyze the impact of derivative videos, resulting in an inability to accurately characterize the actual dissemination effect of the target film and television works, the purpose of the present invention is to provide a method for evaluating the overseas dissemination effect of film and television works based on a multimodal large model. The technical solution adopted is as follows:

[0006] The present invention proposes a method for evaluating the overseas dissemination effect of film and television works based on a multimodal large model, the method comprising:

[0007] Obtain the initial evaluation effect of the target film and television work in the current evaluation cycle according to the four-degree evaluation method;

[0008] Based on the proportion of evaluation phrases on the online platform, the audience preference weight of each evaluation phrase of the target film and television work is obtained, and the evaluation phrases are screened to obtain the audience preference feature vector; the derived evaluation feature vector and its derived evaluation weight of each derivative video are obtained using the same method;

[0009] Matching the audience tendency feature vector with the derived evaluation feature vector, obtaining the guidance emphasis of each derived evaluation feature vector based on the difference between the derived evaluation weight and the audience tendency weight between the two matched vectors, and screening out the derived guidance vector;

[0010] Performing multimodal analysis on the derived video to extract text content and obtain a theme content vector; matching the derived guide vector with the theme content vector and obtaining a guide score for the derived video based on the degree of matching; and screening a guide theme vector from the theme content vector based on the degree of matching;

[0011] The same guiding topic vectors of all derived videos are counted, and the impression tendency score of each guiding topic vector is obtained according to the matching degree of the guiding topic vectors in the matching process of all derived videos and the guiding score; the impression tendency scores of all guiding topic vectors are counted to obtain the derivative video impact score; in the current evaluation cycle, the final evaluation effect is obtained according to the derivative video impact score and the initial evaluation effect.

[0012] Furthermore, the method for obtaining the audience tendency feature vector includes:

[0013] For each evaluation phrase of the target film or television work, the TF-IDF score of each evaluation phrase is calculated, each evaluation phrase is sorted in descending order based on the TF-IDF score, and the sorted sequence number is used as the evaluation level; the audience preference weight of each evaluation phrase of the target film or television work is obtained based on the appearance ratio of the evaluation phrase on the network platform and the evaluation level; and the audience preference feature vector of the target film or television work is obtained based on the evaluation phrases whose audience preference weight is greater than a preset first threshold.

[0014] Furthermore, the method for obtaining the guidance emphasis includes:

[0015] For each derived evaluation feature vector, the ratio of the derived evaluation weight to the audience tendency weight of the matched audience tendency feature vector is normalized to obtain the guidance emphasis of each derived evaluation feature vector.

[0016] Furthermore, the method for obtaining the subject content vector includes:

[0017] For each derivative video, the derivative video is truncated into multiple video judgments based on inter-frame differences, the audio data of each video clip is converted into text, and the text information of each video clip is obtained. The text information is input into the VATT model and multiple topic content vectors are output.

[0018] Furthermore, matching the derived guidance vector with the topic content vector includes:

[0019] For any topic content vector, the absolute value of cosine similarity is calculated between the topic content vector and all derived guide vectors, and the derived guide vector with the largest absolute value of cosine similarity is selected as the derived guide vector matching the topic content vector.

[0020] Furthermore, the method for obtaining the guidance score includes:

[0021] For a set of matched topic content vectors and derived guidance vectors, the absolute value of the cosine similarity in the matching process is multiplied by the guidance emphasis of the derived guidance vector to obtain a weighted vector similarity; the weighted vector similarities of all topic content vectors of the derived video are averaged to obtain the guidance score.

[0022] Furthermore, the method for screening the guided topic vector includes:

[0023] If the absolute value of the cosine similarity between the topic content vector and the matched derived guiding vector is greater than a preset second threshold, the corresponding topic content vector is used as the guiding topic vector.

[0024] Furthermore, the method for obtaining the impression tendency score includes:

[0025] For a guiding topic vector, the weighted vector similarity corresponding to the guiding topic vector in each derivative video is multiplied by the guiding score of the derivative video to obtain the initial impact score of the guiding topic vector in each derivative video; the initial impact scores in all derivative videos are summed and normalized to obtain the impression tendency score of the guiding topic vector.

[0026] Furthermore, the method for obtaining the derived video impact score includes:

[0027] The average impression score of all guided topic vectors is taken as the derived video impact score.

[0028] Furthermore, obtaining the final evaluation result based on the derived video impact score and the initial evaluation result includes:

[0029] For the current evaluation cycle, the derived video impact score is used as the weight of the initial evaluation effect; the negative correlation of the derived video impact score is mapped to obtain the weight of the final evaluation effect of the previous evaluation cycle; the final evaluation effect of the previous evaluation cycle and the initial evaluation effect of the current evaluation cycle are weighted and summed to obtain the final evaluation effect of the current evaluation cycle.

[0030] The present invention has the following beneficial effects:

[0031] The embodiment of the present invention first evaluates a film or television work based on existing processed viewing data to obtain an initial evaluation result. The impact of derivative videos is further analyzed. The present invention uses the target film or television work and the derivative videos as analysis objects, determining the feature vectors and corresponding weights of each object based on the evaluation information. A matching method is then used to determine the guidance emphasis of each derivative evaluation feature vector in the derivative video. The guidance emphasis represents the degree to which the evaluation information corresponding to the derivative evaluation feature vector guides the existing evaluation of the target film or television work. A higher guidance emphasis indicates that the evaluation information deepens the audience's understanding and impression of the target film or television work, thereby screening the derivative guidance vectors of the derivative videos. Considering that the extended video is also a form of multimodal information, a multimodal analysis is performed on it to extract text content and extract the theme content vector. A similar matching operation is performed. Based on the degree of matching, the degree of influence of the theme content of the derivative video on the evaluation information of the derived guidance vector can be evaluated. That is, the higher the degree of matching, the more the content of the derivative video guides and promotes its evaluation information. Because the derived guidance vector is a high-matching information of the audience's tendency feature vector, the guidance effect of the derivative video on the target film or television data can be characterized by a guidance score. Similarly, the theme content vector can be further screened based on the degree of matching to determine the guiding theme vector. Because different derivative videos may have the same theme content, the matching degree and guidance score of the same derivative theme vector in all derivative videos can be counted to determine the impression tendency score of each guiding theme vector. The larger the impression tendency score, the greater the influence of the theme content on the target film and television coordinates on the overseas platform, the stronger the guidance, and the stronger the publicity ability. Therefore, the derivative video influence score can be obtained by statistics and the initial evaluation effect can be adjusted to obtain the final evaluation effect including the viewing information of the target film and television work and the derivative video influence information. The present invention quantifies the influence of the theme content in the derivative video through the information matching method, and obtains the final evaluation effect that can accurately characterize the dissemination effect of the film and television work. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 This is a flow chart of a method for evaluating the overseas dissemination effect of film and television works based on a multimodal large model, provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation method, structure, features and effects of a method for evaluating the overseas dissemination effect of film and television works based on a multimodal large model proposed by the present invention. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics of one or more embodiments may be combined in any suitable form.

[0035] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0036] The following describes in detail a method for evaluating the overseas dissemination effect of film and television works based on a multimodal large model provided by the present invention in conjunction with the accompanying drawings.

[0037] See also Figure 1 , which shows a flow chart of a method for evaluating the overseas dissemination effect of film and television works based on a multimodal large model provided by an embodiment of the present invention, the method comprising:

[0038] Step S1: Obtain the initial evaluation effect of the target film or television work in the current evaluation cycle according to the four-dimensional evaluation method.

[0039] This embodiment of the present invention aims to evaluate the influence of derivative videos of a target film or television work on overseas platforms, and then, based on viewing information of the target film or television work, determine the final evaluation effect of the target film or television work's overseas dissemination. Therefore, this embodiment of the present invention first uses the four-dimensional evaluation method to obtain the initial evaluation effect of the target film or television work in the current evaluation cycle.

[0040] It should be noted that the four-dimensional evaluation method for evaluating the evaluation effect of film and television works is a technical means well known to those skilled in the art. The specific method and principle will not be described in detail. The embodiment of the present invention only briefly describes its implementation process:

[0041] (1) Based on the relevant data of the target film and television works on various platforms, a number of indicators such as the degree of dissemination, friendliness, influence, interaction, playback volume, and number of reviews of the target film and television works are constructed;

[0042] The popularity reflects the exposure and reach of the content of a film or TV show to users. It is obtained through a comprehensive analysis of data such as the number of plays or clicks on various platforms, video downloads, and search engine searches for the target film or TV show, such as ranking and relative popularity.

[0043] Friendliness reflects the user attitudes towards the content of a work, and is obtained by calculating the average ratio of positive and negative opinions of users on various platforms regarding the target film or television work (e.g., the ratio of positive reviews to the total number of reviews);

[0044] Influence focuses on reflecting the influence of the media involved in the communication, and is obtained through a comprehensive analysis of the media's importance, number of social media followers, and the activity level of the media account, such as the relative ranking of industry reputation.

[0045] Interaction is an indicator that measures the degree of audience involvement in information dissemination. It is obtained through the forwarding, likes, collections, and rewards of the target film and television works by various media and through the rankings on various platforms.

[0046] (2) The Delphi method and the analytic hierarchy process were used to assign weights to each indicator. That is, a number of experts in related fields such as film and television communication scholars and data analysts were selected to collect their opinions and judge which indicators needed to be added, deleted or merged based on their experience. The importance of each indicator was scored to obtain the indicator system and the subjective weight of each indicator. After the Delphi method determined the indicator system, the AHP analytic hierarchy process was used to further refine the relative weights of each indicator.

[0047] (3) It should be noted that the four-level evaluation method has four first-level indicators and numerous second- and third-level indicators. The indicators at each level are obtained by weighting the indicators at the next level. Here, only the process of constructing the communication effect score through the first-level indicators is shown. The indicator values ​​corresponding to each indicator are weighted according to their relative weights to obtain the target film and television work in the current evaluation cycle. Communication effect score on :

[0048]

[0049] Where, Indicates that in the current period Next The indicator value corresponding to the indicator; Indicates the The relative weights of the indicators. Indicates the total number of indicators.

[0050] Step S2: Based on the appearance ratio of the evaluation phrases on the network platform, the audience preference weight of each evaluation phrase of the target film and television work is obtained and the evaluation phrases are screened to obtain the audience preference feature vector; the derived evaluation feature vector and its derived evaluation weight of each derivative video are obtained according to the same method.

[0051] Due to different language environments in different regions, when overseas audiences watch the same film or television work, there will be some differences in the ratings, comments, barrages, etc., that is, when watching film or television works, local audiences will prefer works or content that conform to the local language environment. Therefore, when analyzing, it is first necessary to determine the local audience's inclination towards different film and television works based on local evaluation data. Based on the distribution of evaluation segmentations in the evaluation data, the characteristic segmentations that can represent the film and television works are determined, and then when analyzing the tendency in the subsequent steps, the local audience's inclination towards different film and television works can be accurately obtained. Since the embodiment of the invention needs to analyze the audience's understanding and impression of the film and television works after being affected by the derivative videos in the subsequent steps, it is necessary to consider the audience's inclination towards the film and television works expressed through the evaluation information. Therefore, in this step, it is necessary to extract the audience's characteristic segmentations and their corresponding weights for the target film and television works and the derivative videos.

[0052] The embodiment of the present invention takes the target film and television work and the derivative video as the processing objects respectively, takes the target film and television work as an example, analyzes all its evaluation information in the current evaluation cycle, and counts the appearance ratio of each evaluation phrase, that is, the more the evaluation phrase appears, the more the content of the phrase represents the audience's main tendency for the target film and television work, which is reflected in the evaluation phrase. Therefore, the audience tendency weight of each evaluation phrase can be obtained based on this, and the evaluation phrases can be screened based on this, and then the audience tendency feature vector can be obtained by screening the evaluation phrases with audience tendency. That is, the audience tendency feature vector is the vector expression form of the evaluation phrase with audience tendency. Based on the same method, the derivative evaluation feature vector and its derivative evaluation weight of each derivative video can be obtained.

[0053] Preferably, in an embodiment of the present invention, taking the audience tendency feature vector as an example, the method for obtaining the audience tendency feature vector includes:

[0054] First, in an embodiment of the present invention, the Jieba word segmentation method is used to segment the text content in the evaluation information of the target film or television work to obtain a number of evaluation phrases.

[0055] For each review phrase for the target film or television work, a TF-IDF score is calculated for each review phrase. Each review phrase is then sorted in descending order based on the TF-IDF score, with the resulting sequence number serving as the rating level. Specifically, a lower rating indicates a higher TF-IDF score for the corresponding review phrase, indicating greater importance of the review phrase in the current corpus. It should be noted that when performing TF-IDF scoring, the corpus is constructed using all review phrases for the target film or television work. The specific scoring method is well known to those skilled in the art and will not be elaborated on here.

[0056] The audience preference weight of each evaluation phrase for the target film or television work is obtained based on the occurrence ratio of the evaluation phrase on the online platform and the evaluation level. That is, the greater the occurrence ratio and the smaller the corresponding evaluation level, the greater the audience preference weight of the evaluation phrase.

[0057] The audience tendency feature vector of the target film or television work is obtained based on the evaluation phrases whose audience tendency weight is greater than a preset first threshold. Each evaluation phrase can be converted into a vector form through existing encoding or vectorization.

[0058] In this embodiment of the present invention, the occurrence percentage and rating are combined. For each rating phrase, the number of occurrences of the phrase in the corpus is used as the numerator, and the sum of the total number of occurrences of the rating phrases and the rating is used as the denominator to obtain the audience preference weight. In this embodiment of the present invention, because the audience preference weight ranges between 0 and 1 after the fusion of the two features, the first threshold can be set to 0.5.

[0059] It should be noted that the derived evaluation feature vector and derived evaluation weight of each derived video can be obtained according to the same method. The processing object is each derived video, and the corpus of each derived video is constructed according to the evaluation information of each derived video. The setting of the first threshold and the specific weight quantization method are the same and will not be described in detail.

[0060] Step S3: Match the audience tendency feature vector with the derived evaluation feature vector, obtain the guidance emphasis of each derived evaluation feature vector and screen out the derived guidance vector based on the difference between the derived evaluation weight and the audience tendency weight between the two matched vectors.

[0061] After watching the target film or television work, viewers will form their own evaluations and opinions about it, which then form evaluation information on online platforms. However, due to cultural differences or other issues, viewers may have a misunderstanding of the content of the work and continue to watch derivative videos to understand the internal logic and cultural core of the target film or television work. After watching the derivative videos, viewers will follow the guidance of the derivative videos and generate new comments about the target film or television work. By comparing the two types of comments, we can understand the influence of the content degraded by the derivative videos.

[0062] Therefore, the embodiment of the present invention matches the audience tendency feature vector with the derived evaluation feature vector, evaluates the consistency between the two evaluation information through matching, and then derives the guidance of the derived video. For the two matched vectors, the derived evaluation weight and the audience tendency weight are both features obtained by the same method, which characterizes the importance of the corresponding feature vector in the evaluation corpus. Therefore, compared with the audience tendency weight, the larger the derived evaluation weight is, the more receptive and guiding the comment information generated by the derived video is when the two evaluation information are matched. It can also be used to guide the audience to generate corresponding comment content under the derived video. Therefore, the guidance emphasis of each derived evaluation feature vector can be obtained based on the difference between the derived evaluation weight and the audience tendency weight between the two matched vectors. Then, based on the guidance emphasis, the derived guidance vector is screened out from the derived evaluation feature vector of the derived video, that is, the derived guidance vector is a derived evaluation feature vector with a stronger guiding ability for the derived video.

[0063] In an embodiment of the present invention, the method of matching the audience tendency feature vector with the derived evaluation feature vector includes:

[0064] For any derived evaluation feature vector, the absolute value of cosine similarity is calculated between the derived evaluation feature vector and all audience preference feature vectors. The audience preference feature vector with the largest absolute value of cosine similarity is selected as the audience preference feature vector that matches the derived evaluation feature vector. It should be noted that the method for calculating cosine similarity between vectors is a well-known technical means in the art and will not be described in detail here.

[0065] Preferably, because the guidance emphasis actually represents the relative size of the derived evaluation weight relative to the audience preference weight, embodiments of the present invention perform normalization on the ratio of the derived evaluation weight to the audience preference weight of the matching audience preference feature vector for each derived evaluation feature vector to obtain the guidance emphasis of each derived evaluation feature vector. Specifically, a larger ratio indicates a greater derived evaluation weight relative to the audience preference weight, indicating a stronger guidance emphasis in the derived video.

[0066] It should be noted that the normalization method in the embodiment of the present invention can be implemented by mapping methods such as sigmoid function and softsign function, and can also be implemented by range normalization method. The normalization method is a technical means well known to those skilled in the art and will not be described in detail.

[0067] In this embodiment of the present invention, since the guidance emphasis is normalized, the guidance emphasis threshold is set to 0.6, and the derived evaluation feature vectors greater than the guidance emphasis threshold are used as derived guidance vectors. That is, each derived video corresponds to a derived guidance vector set.

[0068] Step S4: Perform multimodal analysis on the derived video to extract text content and obtain a theme content vector; match the derived guidance vector with the theme content vector and obtain a guidance score for the derived video based on the degree of matching; and filter out a guidance theme vector from the theme content vector based on the degree of matching.

[0069] Derivative videos are also multimodal information, encompassing multiple dimensions of data, including images, sound, and text. Therefore, we can perform multimodal analysis on these videos to extract their text content and obtain a topic content vector. This vector is a vector formed by the segmentation of the text within the derived video, and it belongs to the same category as the audience preference feature vector and the derived evaluation feature vector.

[0070] The theme content vector represents the text content in the derivative video, and the above steps are all for analyzing and obtaining evaluation information. Therefore, the embodiment of the present invention further matches the theme content vector with the derivative guidance vector. The greater the degree of matching, the more the derivative video conveys the idea through the corresponding theme content vector, and the audience accepts the theme content and generates similar evaluation information. That is, the greater the degree of matching, the greater the guidance score generated by the derivative video content. Further, similar to the guidance emphasis, the guidance theme vector can be screened out from all theme contents based on the degree of matching. The guidance theme vector represents a vector with strong guidance and audience acceptance among all theme content vectors. After the guidance theme vector is screened out, all derivative videos can be analyzed in the subsequent steps to determine the impact of the theme content on the dissemination of the target film and television work.

[0071] Preferably, in an embodiment of the present invention, the method for obtaining the subject content vector includes:

[0072] For each derivative video, the derivative video is truncated into multiple video judgments based on inter-frame differences, the audio data of each video clip is converted into text, and the text information of each video clip is obtained. The text information is input into the VATT model and multiple topic content vectors are output.

[0073] Similar to the matching process in step S3, matching the derived guidance vector with the topic content vector includes:

[0074] For any topic content vector, the absolute value of cosine similarity is calculated between the topic content vector and all derived guide vectors, and the derived guide vector with the largest absolute value of cosine similarity is selected as the derived guide vector matching the topic content vector.

[0075] It should be noted that the similarity in the two matching processes is the absolute value of the cosine similarity used, because the present invention takes into account the evaluation of dissemination. Although there may be completely different evaluation information, or the subject content is completely different from the evaluation information, the dissemination effect still exists. Whether it is positive publicity that fits the mainstream evaluation of the target film and television work or negative publicity that does not fit, it will guide the audience to think about the content of the work. Therefore, the absolute value of the cosine similarity is used in the two matching processes.

[0076] Preferably, in an embodiment of the present invention, the method for obtaining a guidance score includes:

[0077] For a set of matched topic content vectors and derived guiding vectors, the absolute value of the cosine similarity during the matching process is multiplied by the guiding emphasis of the derived guiding vector to obtain weighted vector similarity. This means that the absolute value of the cosine similarity is weighted using the guiding emphasis as the weight. This weighted vector similarity represents the degree of matching between a set of vectors. A greater guiding emphasis indicates a greater influence on the topic content represented by this matching degree.

[0078] The weighted vector similarities of all the subject content vectors of the derived videos are averaged to obtain the guided score.

[0079] Preferably, in an embodiment of the present invention, the method for screening the guided topic vector includes:

[0080] If the absolute value of the cosine similarity between the topic content vector and the matched derived guidance vector exceeds a preset second threshold, the corresponding topic content vector is used as the guidance topic vector. In this embodiment of the present invention, the second threshold is set to 0.6. As described in the matching process above, the guidance topic vector is content that is completely positively correlated or completely negatively correlated with the evaluation information that produces the guidance effect. This screening method can further avoid evaluation inaccuracies.

[0081] Step S5: Count the same guiding topic vectors of all derived videos, and obtain the impression tendency score of each guiding topic vector based on the matching degree of the guiding topic vectors in the matching process of all derived videos and the guiding score; count the impression tendency scores of all guiding topic vectors to obtain the derivative video impact score; in the current evaluation cycle, obtain the final evaluation effect based on the derivative video impact score and the initial evaluation effect.

[0082] Because different creators have different ways of thinking and knowledge ranges, they have different understandings when watching the target film and television works. This further leads to differences in the creators' understanding of the target film and television works output by many audiences when editing, explaining the target film and television works, and creating derivative videos. Therefore, by integrating the cultural core of the understanding of the target film and television works explained in many derivative videos of the same period, we can derive the core of the target film and television works that are generally accepted and understood locally during that period.

[0083] When watching a film or TV show, audiences often have difficulty understanding the film's content due to cultural conflicts. They will spontaneously watch numerous derivative videos to learn about related content, and then gradually understand the film or TV show. During this process, the audience's acceptance of the film or TV show changes over three stages. At the same time, with the iteration of derivative videos, the audience's evaluation also shows three major stages: initially, negative evaluations caused by cultural conflicts focus on difficult-to-understand aspects such as "confusing plot logic" and "unclear mission motivation." Then, with the numerous derivative videos that fill in the gaps in cultural differences, the quality of these derivative videos varies, causing film reviews to shift from a single difficult-to-understand aspect to many deeper aspects with mixed reviews. Finally, with the fermentation of many good derivative videos and the gradual increase in the number of viewers who understand the core of the film or TV show, the audience's understanding of the work will gradually become unified, causing their film reviews to shift to a unified focus on several cultural content directions.

[0084] When a film or television work is disseminated, derivative videos from different periods of time can spark discussion about the work, thereby increasing its dissemination. When the core content of the target work, as derived from the derivative video's discussion, is relatively consistent with local preferences for the work, local audiences will more quickly accept the derivative video's content and promote discussion of the derivative video. This derivative video, in turn, draws audience attention to the work, thereby spreading the target work. Therefore, the dissemination effect of a film or television work is influenced by its derivative videos. Therefore, embodiments of the present invention calculate the same guiding topic vectors for all derivative videos. Based on the matching degree of the guiding topic vectors across all derivative videos and the guiding scores, an impression tendency score for each guiding topic vector is obtained. By calculating the information represented by the guiding topic vectors during the analysis of different derivative videos, the influence of the guiding topic vector can be quantified, thereby obtaining an impression tendency score. Further, the impression tendency scores of all guiding topic vectors are calculated to obtain a derivative video impact score. This score represents the promotional impact of all derivative videos on the platform for the target work during the current evaluation period, reflecting the extent of the work's dissemination and the contribution of the derivative videos to the promotion and dissemination of the target work.

[0085] It should be noted that because the operations performed in this embodiment of the present invention are all generated within a single evaluation cycle, the derived video impact score obtained is the same as the initial evaluation result, and both are data from the current evaluation cycle. Therefore, in the current evaluation cycle, the final evaluation result is obtained based on the derived video impact score and the initial evaluation result.

[0086] Preferably, in an embodiment of the present invention, the method for obtaining the impression tendency score includes:

[0087] For a given guiding theme vector, the weighted vector similarity corresponding to the guiding theme vector in each derivative video is multiplied by the derivative video's guiding score to obtain the initial influence score of the guiding theme vector in each derivative video. The guiding score is used as a weight. A higher guiding score indicates a better guiding effect for the derivative video, and the influence score calculated by the guiding theme vector during the analysis of the derivative video should be greater.

[0088] The initial impact scores of all derived videos are summed and normalized to obtain the impression tendency score of the guiding topic vector. The normalization method has been described above and will not be repeated or limited here.

[0089] In the embodiment of the present invention, the average impression score of all guided topic vectors is used as the derived video impact score.

[0090] Preferably, the embodiment of the present invention takes into account that in the process of evaluating the dissemination effect of the target film and television work overseas, all dissemination stages are treated equally, that is, when evaluating the dissemination of the work, it does not pay attention to the influence of the derivative videos at the current stage on the dissemination of the target film and television work. In fact, when the film and television work is disseminated, the derivative videos will improve the audience's understanding of the target film and television work and will lead to better local dissemination of the target film and television work. This leads to inaccurate evaluation of the dissemination effect directly through current evaluation searches and other situations. That is, the dissemination effect of the target film and television work is not only the current dissemination impact, but also the audience's past evaluation of the target film and television work and the derivative videos will contribute to the current dissemination effect. But in general, the greater the impact score of the current derivative video on the promotion of the target film and television work, the more objective and positive the audience's current evaluation of the work is, and it has greater dissemination potential. Therefore, the final evaluation effect is obtained based on the derivative video impact score and the initial evaluation effect, including:

[0091] For the current evaluation cycle, the derived video impact score is used as the weight of the initial evaluation effect. The negative correlation of the derived video impact score is mapped to obtain the weight of the final evaluation effect of the previous evaluation cycle. The final evaluation effect of the previous evaluation cycle is weighted and summed with the initial evaluation effect of the current evaluation cycle to obtain the final evaluation effect of the current evaluation cycle. In other words, the larger the derived video impact score in the current evaluation stage, the more it is necessary to use the evaluation value of the current evaluation stage as the final evaluation effect; otherwise, it means that the results of the previous evaluation cycle should be used more.

[0092] It should be noted that, because the derived video impact score is a normalized data, the negative correlation mapping result can be obtained by directly subtracting the derived video impact score from 1.

[0093] It should be noted that since there is no previous evaluation cycle for the first evaluation cycle, its final evaluation effect is the multiplication result of the derived video impact score and the initial evaluation effect.

[0094] By applying the same method to each evaluation stage, the dissemination effects of works at different stages can be compared. Similarly, by applying the same evaluation method to different film and television works, the dissemination effects of different film and television works at the same stage can be compared. This allows for an effective evaluation of the dissemination effects of film and television works on overseas platforms.

[0095] In summary, the present invention evaluates film and television works based on the existing processed viewing data to obtain an initial evaluation effect. The guiding emphasis of each derived evaluation feature vector in the derivative video is determined by a matching method, and the derivative guiding vector of the derivative video can be screened out. The theme content vector of the derivative video is extracted and matched with the derived guiding vector, and the guiding theme vector and the guiding score are determined according to the degree of matching. The matching degree and guiding score of the same derivative theme vector in all derivative videos are counted to determine the impression tendency score of each guiding theme vector, obtain the derivative video influence score and adjust the initial evaluation effect. The present invention quantifies the influence of the theme content in the derivative video through the information matching method, and obtains the final evaluation effect that can accurately characterize the dissemination effect of the film and television works.

[0096] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0097] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A method for evaluating the overseas communication effect of film and television works based on a multimodal large model, characterized by: The method comprises: Obtain the initial evaluation effect of the target film and television work in the current evaluation cycle according to the four-dimensional evaluation method; Based on the proportion of evaluation phrases on the online platform, the audience preference weight of each evaluation phrase of the target film and television work is obtained, and the evaluation phrases are screened to obtain the audience preference feature vector; the derived evaluation feature vector and its derived evaluation weight of each derivative video are obtained using the same method; Matching the audience tendency feature vector with the derived evaluation feature vector, obtaining the guidance emphasis of each derived evaluation feature vector based on the difference between the derived evaluation weight and the audience tendency weight between the two matched vectors, and screening out the derived guidance vector; Performing multimodal analysis on the derived video to extract text content and obtain a theme content vector; matching the derived guide vector with the theme content vector and obtaining a guide score for the derived video based on the degree of matching; and screening a guide theme vector from the theme content vector based on the degree of matching; The same guiding topic vectors of all derived videos are counted, and the impression tendency score of each guiding topic vector is obtained according to the matching degree of the guiding topic vectors in the matching process of all derived videos and the guiding score; the impression tendency scores of all guiding topic vectors are counted to obtain the derivative video impact score; in the current evaluation cycle, the final evaluation effect is obtained according to the derivative video impact score and the initial evaluation effect.

2. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 1 is characterized in that: The method for obtaining the audience tendency feature vector includes: For each evaluation phrase of the target film or television work, the TF-IDF score of each evaluation phrase is calculated, each evaluation phrase is sorted in descending order based on the TF-IDF score, and the sorted sequence number is used as the evaluation level; the audience preference weight of each evaluation phrase of the target film or television work is obtained based on the appearance ratio of the evaluation phrase on the network platform and the evaluation level; and the audience preference feature vector of the target film or television work is obtained based on the evaluation phrases whose audience preference weight is greater than a preset first threshold.

3. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 1 is characterized in that: The method for obtaining the guidance emphasis includes: For each derived evaluation feature vector, the ratio of the derived evaluation weight to the audience tendency weight of the matched audience tendency feature vector is normalized to obtain the guidance emphasis of each derived evaluation feature vector.

4. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 1 is characterized in that: The method for obtaining the subject content vector includes: For each derivative video, the derivative video is truncated into multiple video judgments based on inter-frame differences, the audio data of each video clip is converted into text, and the text information of each video clip is obtained. The text information is input into the VATT model and multiple topic content vectors are output.

5. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 1 is characterized in that: Matching the derived guidance vector with the topic content vector includes: For any topic content vector, the absolute value of cosine similarity is calculated between the topic content vector and all derived guide vectors, and the derived guide vector with the largest absolute value of cosine similarity is selected as the derived guide vector matching the topic content vector.

6. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 5 is characterized in that: The method for obtaining the guidance score includes: For a set of matched topic content vectors and derived guidance vectors, the absolute value of the cosine similarity in the matching process is multiplied by the guidance emphasis of the derived guidance vector to obtain a weighted vector similarity; the weighted vector similarities of all topic content vectors of the derived video are averaged to obtain the guidance score.

7. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 6 is characterized in that: The method for screening the guided topic vector includes: If the absolute value of the cosine similarity between the topic content vector and the matched derived guiding vector is greater than a preset second threshold, the corresponding topic content vector is used as the guiding topic vector.

8. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 6 is characterized in that: The method for obtaining the impression tendency score includes: For a guiding topic vector, the weighted vector similarity corresponding to the guiding topic vector in each derivative video is multiplied by the guiding score of the derivative video to obtain the initial impact score of the guiding topic vector in each derivative video; the initial impact scores in all derivative videos are summed and normalized to obtain the impression tendency score of the guiding topic vector.

9. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 1 is characterized in that: The method for obtaining the derivative video impact score includes: The average impression score of all guided topic vectors is taken as the derived video impact score.

10. The method for evaluating the overseas communication effect of film and television works based on a multimodal large model according to claim 1 is characterized in that: The final evaluation result is obtained based on the derived video impact score and the initial evaluation result, including: For the current evaluation cycle, the derived video impact score is used as the weight of the initial evaluation effect; the negative correlation of the derived video impact score is mapped to obtain the weight of the final evaluation effect of the previous evaluation cycle; the final evaluation effect of the previous evaluation cycle and the initial evaluation effect of the current evaluation cycle are weighted and summed to obtain the final evaluation effect of the current evaluation cycle.

Images