Barrage identification method, device and equipment, storage medium and program product

Abstract

The application relates to the technical field of video processing, and provides a barrage identification method and device, equipment, a storage medium and a program product, which can improve the identification accuracy of spoiler barrages, and the method comprises the following steps: obtaining a target barrage which is synchronously displayed with a target video segment; determining a subsequent video segment which has a playing sequence after the target video segment; taking a comparison result between a plot feature of the target barrage and a plot feature of the subsequent video segment as a plot correlation degree between the target barrage and the subsequent video segment; obtaining a barrage segment formed by barrages which are synchronously displayed with the subsequent video segment; taking a comparison result between a semantic feature of the target barrage and a semantic feature of the barrage segment as a semantic correlation degree between the target barrage and the barrage segment; and identifying the target barrage based on the plot correlation degree and the semantic correlation degree, so as to obtain an identification result of whether the target barrage is a spoiler barrage.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of video processing, in particular to a method and device for identifying bullet screen, computer equipment, storage medium and computer program product. BACKGROUND

[0002] With the rapid development of computer technology and intelligent terminals, videos are deeply loved by users because of their ability to quickly and intuitively convey information, and various video applications are also numerous. In some scenarios, users can post bullet screens while watching videos. For example, user Alice can post a bullet screen for a video segment while watching the video or the video segment. Thereafter, when user Bob watches the video segment, the bullet screen posted by user Alice will be presented to user Bob synchronously with the video segment, thereby increasing the interactivity between users.

[0003] Due to the openness of bullet screen content, some bullet screens posted by users are spoiler bullet screens that reveal the plot of subsequent video segments. The synchronous presentation of such bullet screens with video segments can seriously affect the viewing experience of users who have not watched the subsequent video segments and is not conducive to the interaction between users. However, in the traditional way, the identification of spoiler bullet screens only relies on the content of the bullet screen itself, which has the problem of low identification accuracy. SUMMARY

[0004] Therefore, it is necessary to provide a method and device for identifying bullet screen, computer equipment, storage medium and computer program product to solve the above technical problems.

[0005] A method for identifying bullet screen, the method comprising:

[0006] obtaining a target bullet screen that is synchronously presented with a target video segment;

[0007] determining a subsequent video segment that has a play order after the target video segment;

[0008] obtaining a comparison result between a plot feature of the target bullet screen and a plot feature of the subsequent video segment as a relevance in plot between the target bullet screen and the subsequent video segment;

[0009] obtaining a comparison result between a semantic feature of the target bullet screen and a semantic feature of a bullet screen segment formed by bullet screens that are synchronously presented with the subsequent video segment as a relevance in semantics between the target bullet screen and the bullet screen segment;

[0010] identifying the target bullet screen based on the relevance in plot and the relevance in semantics to obtain an identification result of whether the target bullet screen is a spoiler bullet screen.

[0011] A barrage identification device, the device comprises:

[0012] A target barrage acquisition module is configured to acquire a target barrage that is displayed synchronously with a target video segment;

[0013] A subsequent video segment acquisition module is configured to determine a subsequent video segment that is in a play order after the target video segment;

[0014] A plot relevance acquisition module is configured to take a comparison result between a plot feature of the target barrage and a plot feature of the subsequent video segment as a plot relevance between the target barrage and the subsequent video segment.

[0015] A semantic relevance acquisition module is configured to acquire a barrage segment formed by barrages that are displayed synchronously with the subsequent video segment, and take a comparison result between a semantic feature of the target barrage and a semantic feature of the barrage segment as a semantic relevance between the target barrage and the barrage segment.

[0016] A spoiler barrage identification module is configured to identify the target barrage based on the plot relevance and the semantic relevance, and obtain an identification result of whether the target barrage is a spoiler barrage.

[0017] In some embodiments, the spoiler barrage identification module is further configured to acquire a tendency degree of a target account in publishing spoiler barrages, the target account is an account that publishes the target barrage; obtain a probability that the target barrage is a spoiler barrage based on a positive correlation between each of the plot relevance, the semantic relevance, and the tendency degree and a barrage spoiler probability; and obtain the identification result of whether the target barrage is a spoiler barrage based on the probability that the target barrage is a spoiler barrage.

[0018] In some embodiments, the spoiler barrage identification module is further configured to acquire a first weight corresponding to the plot relevance and the semantic relevance, and a second weight corresponding to the tendency degree; and obtain the probability that the target barrage is a spoiler barrage according to a product result between the first weight, the plot relevance, and the semantic relevance, and a product result between the second weight and the tendency degree.

[0019] In some embodiments, the spoiler barrage identification module is further configured to determine a publishing time of the target barrage; determine a total number of barrages published by the target account and having a publishing time before the publishing time of the target barrage; determine a number of spoiler barrages published by the target account and having a publishing time before the publishing time of the target barrage; and take a ratio of the number of spoiler barrages to the total number as the tendency degree of the target account in publishing spoiler barrages.

[0020] In some embodiments, the plot feature of the target barrage is a barrage plot depth representation, and the plot feature of the subsequent video segment is a video segment plot depth representation; the plot relevance obtaining module is further configured to: obtain the barrage plot depth representation of the target barrage, and obtain the video segment plot depth representation of the subsequent video segment; and perform interactive fusion on the barrage plot depth representation and the video segment plot depth representation, and take the interactive fusion result as the plot relevance between the target barrage and the subsequent video segment.

[0021] In some embodiments, the plot relevance obtaining module is further configured to:

[0022] perform speech recognition on the dialogue of the subsequent video segment to obtain dialogue text of the subsequent video segment; and construct the video segment plot depth representation of the subsequent video segment based on the dialogue text.

[0023] and / or,

[0024] perform character recognition on the video frames of the subsequent video segment to obtain subtitle text of the subsequent video segment; and construct the video segment plot depth representation of the subsequent video segment based on the subtitle text.

[0025] In some embodiments, the semantic feature of the target barrage is a barrage semantic depth representation, and the semantic feature of the barrage segment is a barrage segment semantic depth representation; the semantic relevance obtaining module is further configured to: obtain the barrage semantic depth representation of the target barrage, and obtain the barrage segment semantic depth representation of the barrage segment; and perform interactive fusion on the barrage semantic depth representation and the barrage segment semantic depth representation, and take the interactive fusion result as the semantic relevance between the target barrage and the barrage segment.

[0026] In some embodiments, the target barrage obtaining module is further configured to: obtain a plurality of barrages that are synchronously displayed with a target video segment; perform plot detection on each barrage in the plurality of barrages respectively to obtain a plot detection result representing whether the barrage has a plot; and determine, according to the plot detection result, a barrage having a plot in the plurality of barrages as a target barrage.

[0027] In some embodiments, the target barrage obtaining module is further configured to: perform word segmentation processing on each barrage to obtain a word segmentation sequence corresponding to each barrage; obtain a plot detection feature corresponding to the barrage according to each word in the word segmentation sequence and a position of each word in the word segmentation sequence; and input the plot detection feature into a pre-trained plot detection neural network to output, by the plot detection neural network, a plot detection result representing whether the barrage has a plot.

[0028] In some embodiments, the plot relevance acquisition module is further used to input the plot features of the target barrage and the plot features of the subsequent video clip into a pre-trained plot relevance detection neural network for comparison; and use the comparison result output by the plot relevance detection neural network as the plot relevance between the target barrage and the subsequent video clip.

[0029] In some embodiments, the device also includes a plot relevance detection neural network training module, which is used to obtain multiple sample barrages displayed synchronously with the sample video clip; determine the first sample barrage and the second sample barrage among the multiple sample barrages; when the audience recognition corresponding to the first sample barrage is higher than the audience recognition corresponding to the second sample barrage, determine that the plot relevance of the first sample barrage to the sample video clip is higher than the plot relevance of the second sample barrage to the sample video clip; based on the fact that the plot relevance of the first sample barrage to the sample video clip is higher than the plot relevance of the second sample barrage to the sample video clip, the plot features of the first sample barrage, the plot features of the second sample barrage and the plot features of the sample video clip, train the plot relevance detection neural network.

[0030] In some embodiments, the device also includes an audience recognition determination module, which is used to obtain the number of likes of the first sample barrage and the number of likes of the second sample barrage; when the number of likes of the first sample barrage is greater than the number of likes of the second sample barrage, it is determined that the audience recognition corresponding to the first sample barrage is higher than the audience recognition corresponding to the second sample barrage.

[0031] A computer device includes a memory and a processor, wherein the memory stores a computer program and the processor executes the above-mentioned bullet comment recognition method.

[0032] A computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the above-mentioned bullet comment recognition method.

[0033] A computer program product includes a computer program, which implements the above-mentioned barrage recognition method when executed by a processor.

[0034] In the above-mentioned barrage recognition method, device, computer equipment, storage medium and computer program product, when it is necessary to identify the target barrage as a spoiler barrage, after obtaining the target barrage displayed synchronously with the target video clip and the subsequent video clip whose playback order is after the target video clip, based on the comparison between the plot features of the target barrage and the plot features of the subsequent video clip, the plot correlation between the target barrage and the subsequent video clip is obtained; based on the comparison between the semantic features of the target barrage and the semantic features of the barrage segment displayed synchronously with the subsequent video, the semantic correlation between the target barrage and the barrage segment displayed synchronously with the subsequent video clip is obtained; then, based on the plot correlation and the semantic correlation, the target barrage is identified to obtain an identification result of whether the target barrage is a spoiler barrage. On the one hand, based on the definition of spoiler barrage, the identification of spoiler barrage does not rely solely on the target barrage itself, but on the correlation between the target barrage and the candidate video clips. On the other hand, the correlation includes two dimensions: plot relevance and semantic relevance, which fully considers the intrinsic connection between the target barrage and the candidate video clips, thereby greatly improving the accuracy of identifying spoiler barrage. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A diagram showing an application environment of a bullet comment recognition method in one embodiment;

[0036] Figure 2 1 is a flow chart of a method for identifying bullet comments in one embodiment;

[0037] Figure 3 Schematic diagram of the flow of a bullet comment recognition method in another embodiment;

[0038] Figure 4 A schematic diagram of a process for obtaining plot relevance in one embodiment;

[0039] Figure 5 A schematic diagram of a process for obtaining semantic relevance in some embodiments;

[0040] Figure 6 A schematic diagram of a process for detecting whether a bullet comment has a plot in some embodiments;

[0041] Figure 7 1 is a flow chart of a method for identifying bullet comments in one embodiment;

[0042] Figure 8 1 is a flow chart of a method for identifying bullet comments in one embodiment;

[0043] Figure 9 This is a structural block diagram of a bullet comment recognition device in one embodiment;

[0044] Figure 10 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0046] References to "embodiments" in this application mean that specific features, structures, or characteristics described in connection with the embodiments may be included in at least some embodiments of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0047] The bullet screen recognition method provided by this application can be applied to Figure 1 In the application environment shown, the terminal 110 can communicate with the server 120 via a network; the terminal 110 obtains a video from the server 120 via the network and plays the video to the user. While the terminal 110 plays the video to the user, it can simultaneously display the barrage posted for the video. Before the terminal 110 synchronously displays the barrage for the video, the server 120 can first use the barrage recognition method provided by the present application to identify spoiler barrages for the barrages displayed synchronously with the current video clip; if the server 120 recognizes that the barrage is a spoiler barrage, it can dynamically process the spoiler barrage based on the user's viewing of subsequent video clips. The dynamic processing is, for example: if the subsequent video clip corresponding to the plot revealed by the spoiler barrage has been watched by the user, the spoiler barrage will not be blocked, and the spoiler barrage will be synchronously displayed to the user along with the current video clip; the dynamic processing can also be: if the subsequent video clip corresponding to the plot revealed by the spoiler barrage has not been watched by the user, the spoiler barrage will be blocked and will not be displayed to the user along with the current video clip.

[0048] In one embodiment, while playing the video to the user, the terminal 110 may also receive a comment posted by the user for the current video segment, and feed the comment posted by the user for the current video segment back to the server 120. The server 120 may use the comment recognition method provided in this application to identify whether the comment posted by the user for the current video segment is a spoiler comment.

[0049] The terminal 110 can be, but is not limited to, various desktop computers, notebook computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle-mounted device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The terminal 110 can be installed with a client, which can be used to play a video. The server 120 can be implemented by an independent server or a server cluster or a cloud server composed of multiple servers.

[0050] In some embodiments, as shown in Figure 2 , a method for identifying a barrage is provided. The method is applied to a computer device (such as the terminal 110 or the server 120) in Figure 1 for example, and includes the following steps:

[0051] In step S201, a target barrage that is to be synchronously displayed with a target video segment is obtained.

[0052] The target barrage is a barrage to be identified as a spoiler barrage. The target video segment is any one of a plurality of video segments obtained by dividing an original video. For example, a video segment to be played by the terminal can be taken as the target video segment, and the computer device obtains a target barrage to be synchronously displayed with the target video segment.

[0053] The barrage can be understood as a comment containing text content that is synchronously displayed with a video. A user can post a barrage for a video being watched during the process of watching the video, and the computer device records the posted barrage in correspondence with the video. The target barrage can be any one of a plurality of barrages synchronously displayed with the target video segment.

[0054] In an embodiment, after the computer device determines the target video segment from a plurality of video segments obtained by video segmentation, the computer device obtains a plurality of barrages synchronously displayed with the target video segment, and the computer device sequentially identifies the plurality of barrages as spoiler barrages to obtain identification results corresponding to the barrages, wherein the barrage processed by the computer device each time is the target barrage.

[0055] In step S202, a subsequent video segment after the target video segment in a playing order is determined.

[0056] The playing order of the video clips is the playing order of the video clips in the original video, and the subsequent video clip is a video clip after the target video clip in the playing order. The number of the subsequent video clip can be one. For example, it is the next video clip after the target video clip. The number of the subsequent video clip can be multiple. For example, the subsequent video clip can be all video clips after the target video clip in the playing order in the original video. It can be understood that the later the playing order of the target video clip in the original video, the fewer the number of the subsequent video clip after the target video clip, and the earlier the playing order of the target video clip in the original video, the more the number of the subsequent video clip after the target video clip.

[0057] Exemplarily, the video clips obtained by segmenting the video are video clip 1, video clip 2, video clip 3 and video clip 4, and if the video clip 2 is the target video clip, since the playing order of the video clip 3 and the video clip 4 are both after the video clip 2, the computer device can select one of the video clip 3 and the video clip 4 as the subsequent video clip. The computer device can select the video clip 4 as the subsequent video clip, and detect whether the video clip 2 is a spoiler bullet screen for the video clip 4. The computer device can also select the video clip 3 as the subsequent video clip, and detect whether the video clip 2 is a spoiler bullet screen for the video clip 3.

[0058] In some embodiments, the subsequent video clip can also be a video clip after the target video clip in the playing order and having a plot association with the target video clip. Generally, the plot described by the target bullet screen is generally related to the target video clip. If the target bullet screen spoils the plot of the subsequent video clip, the target bullet screen is also related to the subsequent video clip, and in this case, to some extent, the subsequent video clip also has a plot association with the target video clip. Specifically, the computer device can store video clips having a plot association with each video clip in advance, obtain a video clip after the target video in the playing order from the video clips, and determine the subsequent video clip.

[0059] In an embodiment, the computer device can also obtain historical playback record data corresponding to the current user identifier, determine a video clip not watched by the user according to the historical playback record data, and filter a video clip not watched by the current user from the video clip after the target video clip in the playing order, so as to determine the subsequent video clip.

[0060] In an embodiment, the computer device can further filter, from the video clips following the target video in the play order, the video clips that have a plot relevance with the target video and have not been watched by the current user, as the subsequent video clips. In this way, the computer device only needs to compare the target barrage with part of the subsequent video clips, instead of all the subsequent video clips, reducing the number of comparisons and improving the recognition efficiency.

[0061] In step S203, the computer device obtains the comparison result between the plot feature of the target barrage and the plot feature of the subsequent video clip as the plot relevance of the target barrage and the subsequent video clip.

[0062] The plot feature of the barrage is a feature related to the video plot embodied by the barrage. The plot feature of the target barrage can be a plot depth representation corresponding to the text content contained in the target barrage.

[0063] The plot feature of the video clip is a feature related to the video plot embodied by the video clip. The plot feature of the subsequent video clip can be a plot depth representation corresponding to the video text content of the subsequent video clip.

[0064] In an embodiment, the computer device can obtain the target barrage, extract the text content from the target barrage, and obtain the corresponding plot feature based on the text content.

[0065] In an embodiment, the computer device can obtain the video text content of the subsequent video clip, and obtain the corresponding plot feature based on the video text content. Illustratively, the video text content can include the dialogue, subtitles, lines, narration or plot introduction of the video, etc. For example, the computer device can obtain the video text content by performing speech recognition on the dialogue of the candidate video clip. For another example, the computer device can obtain the video text content by performing optical character recognition on the video frames in the subsequent video clip.

[0066] The plot relevance of the target barrage and the subsequent video clip is the plot relevance of the target barrage and the subsequent video clip. The greater the plot relevance of the target barrage and the subsequent video clip, the more relevant the plot described by the target barrage and the plot described by the subsequent video clip.

[0067] Specifically, after obtaining the plot feature of the target barrage and the plot feature of the subsequent video clip, the computer device performs similarity comparison on the target barrage and the plot feature and the plot feature of the subsequent video clip. The more similar the plot feature of the target barrage and the plot feature of the subsequent video clip, the higher the plot relevance of the target barrage and the subsequent video clip. This plot relevance can be referred to as plot relevance.

[0068] In step S204, the barrage segment formed by the barrages synchronously displayed with the subsequent video segment is obtained, and the comparison result between the semantic feature of the target barrage and the semantic feature of the barrage segment is taken as the semantic correlation degree between the target barrage and the barrage segment.

[0069] The barrage segment is a text paragraph formed by the text content included in the multiple barrages. Specifically, the computer device can take the text content included in the multiple barrages synchronously displayed with the subsequent video segment as the barrage segment, and the publishing time of each barrage included in the barrage segment can be earlier than the publishing time of the target barrage or later than the publishing time of the target barrage. For example, the computer device can splice the text content included in each barrage synchronously displayed with the subsequent video segment to obtain a text paragraph as the barrage segment.

[0070] The semantic feature of the barrage is a feature related to the text semantics embodied by the barrage. The semantic feature of the target barrage can be a semantic depth representation corresponding to the text content included in the target barrage.

[0071] The semantic feature of the barrage segment is a feature related to the text semantics embodied by the barrage segment. The barrage segment includes multiple barrages, and each barrage includes text content. For example, the semantic feature of the barrage segment can be determined based on the semantic features of the barrages included in the barrage segment. For example, the computer device can obtain a semantic depth representation corresponding to the text content of each barrage included in the barrage segment, and obtain a semantic depth representation corresponding to the barrage segment based on the semantic depth representation of each barrage. For example, the semantic feature of the barrage segment can be directly determined based on the text paragraph of the barrage segment itself. For example, the computer device obtains the barrage segment, that is, the text paragraph, and directly takes the semantic depth representation corresponding to the entire text paragraph as the semantic depth representation corresponding to the barrage segment.

[0072] The semantic correlation degree between the target barrage and the barrage segment is the semantic correlation between the target barrage and the barrage segment. The greater the semantic correlation degree between the target barrage and the barrage segment, the more similar the semantics embodied by the target barrage and the semantics embodied by the barrage segment.

[0073] Specifically, after obtaining the semantic feature of the target barrage and the semantic feature of the barrage segment of the subsequent video segment, the computer device performs similarity comparison on the semantic feature of the target barrage and the semantic feature of the barrage segment of the subsequent video segment. The more similar the semantic feature of the target barrage and the semantic feature of the barrage segment of the subsequent video segment, the higher the semantic correlation degree between the target barrage and the barrage segment of the subsequent video segment. The semantic correlation degree can be referred to as a semantic correlation degree.

[0074] In step S205, the target barrage is identified based on the plot correlation degree and the semantic correlation degree, and an identification result of whether the target barrage is a spoiler barrage is obtained.

[0075] The plot relevance is positively correlated with the probability that the target barrage is a spoiler barrage. That is, the higher the plot relevance, the greater the probability that the target barrage is a spoiler barrage. The semantic relevance is positively correlated with the probability that the target barrage is a spoiler barrage. That is, the higher the semantic relevance, the greater the probability that the target barrage is a spoiler barrage.

[0076] In one embodiment, after determining the plot relevance between the target barrage and the subsequent video segment, as well as the semantic relevance between the target barrage and the barrage segments of the subsequent video segment, the computer device determines the probability that the target barrage is a spoiler barrage based on the positive correlation described above, completes spoiler identification of the target barrage, and uses the probability of the target barrage being a spoiler barrage as the identification result. If the probability of the target barrage being a spoiler barrage is greater than or equal to a probability threshold, the target barrage is determined to be a spoiler barrage; if the probability of the target barrage being a spoiler barrage is less than the probability threshold, the target barrage is determined not to be a spoiler barrage.

[0077] In the above-mentioned barrage recognition method, when it is necessary to identify a target barrage as a spoiler barrage, after obtaining the target barrage displayed synchronously with the target video clip and the subsequent video clip whose playback order is after the target video clip, the plot relevance between the target barrage and the subsequent video clip is obtained based on the comparison between the plot features of the target barrage and the plot features of the subsequent video clip; the semantic relevance between the target barrage and the barrage segment displayed synchronously with the subsequent video clip is obtained based on the comparison between the semantic features of the target barrage and the semantic features of the barrage segment displayed synchronously with the subsequent video clip; then, based on the plot relevance and the semantic relevance, the target barrage is identified to obtain the identification result of whether the target barrage is a spoiler barrage. On the one hand, based on the definition of spoiler barrage, it does not rely solely on the target barrage itself, but relies on the relevance between the target barrage and the candidate video clip to identify spoiler barrage. On the other hand, the relevance includes two dimensions: plot relevance and semantic relevance, which fully considers the intrinsic connection between the target barrage and the candidate video clip, thereby greatly improving the accuracy of identifying spoiler barrage.

[0078] In some embodiments, the computer device can determine whether the target barrage is a spoiler barrage based on the plot relevance between the target barrage and the subsequent video clip, the semantic relevance between the target barrage and the barrage segment of the subsequent video clip, and the tendency of the account that posts the target barrage to post spoiler barrages.

[0079] Specifically, the computer device can obtain the target account's tendency to post spoiler barrages; the target account is the account that posts the target barrage; based on the positive correlation between the plot relevance, semantic relevance, and tendency and the probability of barrage spoilers, the probability that the target barrage is a spoiler barrage is obtained; based on the probability that the target barrage is a spoiler barrage, the identification result of whether the target barrage is a spoiler barrage is obtained.

[0080] An account's propensity to post spoiler comments can be calculated by calculating the ratio of the number of spoiler comments posted by the account to the total number of comments posted by the account. The higher the ratio of the number of spoiler comments posted by the account to the total number of comments posted by the account, the higher the account's propensity to post spoiler comments.

[0081] The target account's tendency to post spoiler comments is positively correlated with the probability that the target comment is a spoiler comment. That is, the higher the target account's tendency to post spoiler comments, the higher the probability that the target comment is a spoiler comment.

[0082] After obtaining the plot relevance between the target barrage and the subsequent video segment, the semantic relevance between the target barrage and the barrage segment of the subsequent video segment, and the tendency of the account that posted the target barrage to post spoiler barrages, the computer device can obtain the probability that the target barrage is a spoiler barrage based on the positive correlation between the plot relevance, semantic relevance, and tendency and the probability of the barrage being a spoiler, and use the probability of the target barrage being a spoiler barrage as the identification result. If the probability that the target barrage is a spoiler barrage is greater than or equal to the probability threshold, the target barrage is determined to be a spoiler barrage. If the probability that the target barrage is a spoiler barrage is less than the probability threshold, the target barrage is determined not to be a spoiler barrage.

[0083] In this embodiment, the target account's tendency to post spoiler comments, the plot relevance of the target comment and the subsequent video clip, and the semantic relevance of the target comment and the comment segments of the subsequent video clip are combined to identify whether the target comment is a spoiler comment, thereby improving the accuracy of recognition.

[0084] In some embodiments, the step of obtaining the target account's tendency to post spoiler barrages includes: determining the release time of the target barrage; determining the total number of barrages posted by the target account and whose release time is before the release time of the target barrage; determining the number of spoiler barrages posted by the target account and whose release time is before the release time of the target barrage; and taking the ratio of the number of spoiler barrages to the total number as the target account's tendency to post spoiler barrages.

[0085] The tendency of the target account to post spoiler bullet screen is denoted as Pu. For example, the time when the target account posts the target bullet screen (i.e., the posting time of the target bullet screen) is t n The computer device can then determine the total number of bullet screens posted by the target account and having a posting time before t n The computer device can then determine the number of spoiler bullet screens posted by the target account and having a posting time before t n The computer device can then determine the number of spoiler bullet screens posted by the target account and having a posting time before t

[0086] In this embodiment, the historical bullet screens posted by the account that posts the target bullet screen are statistically analyzed. The ratio of the number of spoiler bullet screens posted by the target account to the total number of historical bullet screens posted by the target account is used as the tendency of the target account to post spoiler bullet screens, which improves the accuracy of the tendency.

[0087] In some embodiments, based on the positive correlation between the tendency and the spoiler bullet screen possibility, the step of obtaining the probability that the target bullet screen is a spoiler bullet screen specifically includes: obtaining a first weight corresponding to the plot-related degree and the semantic-related degree, and a second weight corresponding to the tendency; and obtaining the probability that the target bullet screen is a spoiler bullet screen according to the product of the first weight, the plot-related degree and the semantic-related degree, and the product of the second weight and the tendency.

[0088] The subsequent video segment is denoted as s, the possibility that the target bullet screen is a spoiler bullet screen for the subsequent video segment is denoted as Pt[s], the plot-related degree between the target bullet screen and the subsequent video segment s is denoted as Pr[s], the semantic-related degree between the bullet screen segment synchronously displayed by the target bullet screen and the subsequent video segment s is denoted as Prd[s], the first weight corresponding to the plot-related degree Pr[s] and the semantic-related degree Prd[s] is denoted as w1, the tendency of the target account to post spoiler bullet screens is denoted as Pu, and the second weight corresponding to the tendency Pu is denoted as w2.

[0089] Since the plot relevance, the semantic relevance, and the tendency degree each have a positive correlation with the probability of the spoiler, w1>0 and w2>0, and w1+w2=1 (i.e., the sum of the first weight and the second weight is 1). w1 and w2 can be the optimal parameters found by grid search on the constructed evaluation data. The product result w1×Pr[s]×Prd[s] between the first weight w1, the plot relevance Pr[s], and the semantic relevance Prd[s], and the product result w2×Pu between the second weight w2 and the tendency degree Pu are added, and the addition result is taken as the possibility that the target barrage is a spoiler barrage, i.e., Pt[s]=w1×Pr[s]×Prd[s]+w2×Pu.

[0090] In this embodiment, the possibility that the target barrage is a spoiler barrage is obtained according to the product result between the first weight, the plot relevance, and the semantic relevance, and the product result between the second weight and the tendency degree, so as to improve the identification accuracy of the spoiler barrage.

[0091] As shown in FIG. 1, it is a flowchart of the barrage identification method in an embodiment. The method can be executed by a computer device. Referring to FIG. 2, the method mainly includes the following steps: Figure 3 Figure 3

[0092] Step S301, obtaining a target barrage that is synchronously displayed with a target video segment;

[0093] Step S302, determining a subsequent video segment that is played after the target video segment;

[0094] Step S303, taking the comparison result between the plot feature of the target barrage and the plot feature of the subsequent video segment as the plot relevance between the target barrage and the subsequent video segment;

[0095] Step S304, obtaining a barrage segment formed by barrages that are synchronously displayed with the subsequent video segment, and taking the comparison result between the semantic feature of the target barrage and the semantic feature of the barrage segment as the semantic relevance between the target barrage and the barrage segment;

[0096] Step S305, determining the total number of barrages published by a target account and published before the publication time of the target barrage, and determining the number of spoiler barrages published by the target account and published before the publication time of the target barrage;

[0097] Step S306, taking the ratio of the number of the spoiler barrages to the total number as the tendency degree of the target account in publishing the spoiler barrages;

[0098] ​​Step S307, obtaining a first weight corresponding to the plot relevance and the semantic relevance, and a second weight corresponding to the tendency;

[0099] Wherein, the sum of the first weight and the second weight is 1;

[0100] Step S308: Obtain the probability that the target bullet comment is a spoiler bullet comment based on the product of the first weight, the plot relevance, and the semantic relevance, and the product of the second weight and the tendency.

[0101] For the introduction of relevant steps, reference may be made to the embodiments mentioned above.

[0102] In this embodiment, the target account's tendency to post spoiler barrages, the plot relevance between the target barrage and subsequent video clips, and the semantic relevance between the target barrage and the barrage segments of subsequent video clips are combined to identify whether the target barrage is a spoiler barrage, thereby improving the accuracy of recognition; and, a statistical analysis is performed on the barrages historically posted by the account that posted the target barrage, and the ratio of the number of spoiler barrages historically posted by the target account to the total number of barrages historically posted is used as the target account's tendency to post spoiler barrages, thereby improving the accuracy of the tendency; in addition, based on the product result of the first weight, the plot relevance and the semantic relevance, and the product result of the second weight and the tendency, the probability that the target barrage is a spoiler barrage is obtained, thereby improving the recognition accuracy of spoiler barrages.

[0103] In some embodiments, based on the comparison between the plot features of the target barrage and the plot features of the subsequent video clip, the plot correlation between the target barrage and the subsequent video clip is obtained, including: obtaining the barrage plot depth representation of the target barrage, obtaining the video clip plot depth representation of the subsequent video clip; interactively fusing the barrage plot depth representation and the video clip plot depth representation, and using the interactive fusion result as the plot correlation between the target barrage and the subsequent video clip.

[0104] Among them, the plot features of the target barrage are the barrage plot depth representation, and the plot features of the subsequent video clips are the video clip plot depth representation.

[0105] like Figure 4 As shown, Figure 4 FIG. 1 is a flow chart of obtaining plot relevance in one embodiment. Figure 4The computer device can input the text of the target barrage, the dialogue text and the subtitle text of the subsequent video segment into the BERT layer (Bidirectional Encoder Representation from Transformers, a semantic representation network based on a neural network) to obtain a barrage scenario depth representation of the target barrage and a video segment scenario depth representation of the subsequent video segment. Then, the barrage scenario depth representation and the video segment scenario depth representation are interactively fused through an attention mechanism layer (Attention) to obtain the relevance of the target barrage and the subsequent video segment in terms of scenarios.

[0106] In the above manner, the interactive fusion of the barrage scenario depth representation of the target barrage and the video segment scenario depth representation of the subsequent video segment can ensure the accuracy of the relevance of the target barrage and the subsequent video segment in terms of scenarios.

[0107] In some embodiments, obtaining the video segment scenario depth representation of the subsequent video segment can include the following steps: performing speech recognition on dialogue in the subsequent video segment to obtain dialogue text of the subsequent video segment; constructing the video segment scenario depth representation of the subsequent video segment based on the dialogue text; and / or performing character recognition on a video frame of the subsequent video segment to obtain subtitle text of the subsequent video segment; and constructing the video segment scenario depth representation of the subsequent video segment based on the subtitle text.

[0108] Specifically, after obtaining the subsequent video segment, the computer device can perform automatic speech recognition (ASR, Automatic Speech Recognition) on dialogue in the subsequent video segment to obtain dialogue text, input the dialogue text into the BERT layer shown in FIG. 2, and obtain the video segment scenario depth representation of the subsequent video segment. Figure 4 In the BERT layer shown in FIG. 2, the video segment scenario depth representation of the subsequent video segment is obtained. The computer device can also extract a video frame (which can be understood as an image frame) from the subsequent video segment, perform optical character recognition (OCR, Optical Character Recognition) on the video frame to obtain subtitle text, and input the subtitle text into the BERT layer shown in FIG. 2 to obtain the video segment scenario depth representation of the subsequent video segment. Figure 4 In the BERT layer shown in FIG. 2, the video segment scenario depth representation of the subsequent video segment is obtained. The computer device can also extract a video frame (which can be understood as an image frame) from the subsequent video segment, perform optical character recognition (OCR, Optical Character Recognition) on the video frame to obtain subtitle text, and input the subtitle text into the BERT layer shown in FIG. 2 to obtain the video segment scenario depth representation of the subsequent video segment. Figure 4 In the BERT layer shown in FIG. 2, the video segment scenario depth representation of the subsequent video segment is obtained.

[0109] In this embodiment, the dialogue text of the subsequent video segment is obtained through speech recognition, and the subtitle text of the subsequent video segment is obtained through character recognition, so that the constructed video segment scenario depth representation can more truly reflect the scenario characteristics of the subsequent video segment.

[0110] In some embodiments, the comparison result between the plot features of the target barrage and the plot features of the subsequent video segment is used as the plot relevance between the target barrage and the subsequent video segment, including: inputting the plot features of the target barrage and the plot features of the subsequent video segment into a pre-trained plot relevance detection neural network for comparison; and using the comparison result output by the plot relevance detection neural network as the plot relevance between the target barrage and the subsequent video segment.

[0111] Among them, the processing flow of the plot relevance detection neural network can be as follows Figure 4 shown.

[0112] After the computer device obtains the text of the target barrage, as well as the dialogue text and subtitle text of the subsequent video clip, it can input the text of the target barrage, as well as the dialogue text and subtitle text of the subsequent video clip, into the BERT layer of the plot relevance detection neural network, to obtain the barrage plot depth representation and the video clip plot depth representation output by the BERT layer. Next, the computer device interactively fuses the barrage plot depth representation and the video clip plot depth representation through the attention mechanism layer of the plot relevance detection neural network, completes the comparison between the barrage plot depth representation and the video clip plot depth representation, and uses the interactive fusion result (i.e., the comparison result) as the plot relevance between the target barrage and the subsequent video clip.

[0113] In this embodiment, the plot features of the target barrage and the plot features of the subsequent video clips are compared through the plot relevance detection neural network to ensure the accuracy of the relevance of the target barrage and the subsequent video clips in terms of plot.

[0114] In some embodiments, the plot relevance detection neural network is constructed by the following steps:

[0115] Acquire multiple sample barrages displayed synchronously with the sample video clip; determine a first sample barrage and a second sample barrage among the multiple sample barrages; when the audience recognition corresponding to the first sample barrage is higher than the audience recognition corresponding to the second sample barrage, determine that the relevance of the first sample barrage to the sample video clip in terms of plot is higher than the relevance of the second sample barrage to the sample video clip in terms of plot; based on the fact that the relevance of the first sample barrage to the sample video clip in terms of plot is higher than the relevance of the second sample barrage to the sample video clip in terms of plot, the plot features of the first sample barrage, the plot features of the second sample barrage, and the plot features of the sample video clip, train a plot relevance detection neural network.

[0116] The audience recognition of the barrage can represent the degree of recognition of the barrage by the audience.

[0117] For example, among multiple sample bullet screens displayed synchronously with a sample video clip, the sample bullet screen a has a higher audience recognition than the sample bullet screen b, and the sample bullet screen a has a higher scenario correlation with the sample video clip than the sample bullet screen b. The computer device can perform model training according to the scenario correlation between the sample bullet screen a and the sample video clip being higher than the scenario correlation between the sample bullet screen b and the sample video clip, the scenario feature of the sample bullet screen a, the scenario feature of the sample bullet screen b, and the scenario feature of the sample video clip, to obtain a scenario correlation detection neural network.

[0118] In this embodiment, the relative levels of the scenario correlations between the sample bullet screens and the sample video clip are determined based on the relative levels of the audience recognitions obtained by the sample bullet screens, to ensure the prediction accuracy of the scenario correlation detection neural network in scenario correlation.

[0119] In some embodiments, the relative levels between the audience recognitions corresponding to the first sample bullet screen and the second sample bullet screen can be determined by the following steps: obtaining the like numbers of the first sample bullet screen and the second sample bullet screen; and determining that the audience recognition corresponding to the first sample bullet screen is higher than the audience recognition corresponding to the second sample bullet screen when the like number of the first sample bullet screen is greater than the like number of the second sample bullet screen.

[0120] After obtaining the like numbers of the first sample bullet screen and the second sample bullet screen, the computer device determines that the audience recognition obtained by the first sample bullet screen is higher than the audience recognition obtained by the second sample bullet screen when the like number of the first sample bullet screen is greater than the like number of the second sample bullet screen.

[0121] In this embodiment, the relative levels between the audience recognitions obtained by the sample bullet screens are determined by the relative sizes between the like numbers of the sample bullet screens, to ensure the accuracy of the relative levels between the scenario correlations of the multiple sample bullet screens with the sample video clip.

[0122] In some embodiments, the comparison result between the semantic feature of the target bullet screen and the semantic feature of the bullet screen segment can be used as the semantic correlation between the target bullet screen and the bullet screen segment, which can include the following steps: obtaining a bullet screen semantic depth representation representing the semantic feature of the target bullet screen, and obtaining a bullet screen segment semantic depth representation representing the semantic feature of the bullet screen segment; and performing interactive fusion on the bullet screen semantic depth representation and the bullet screen segment semantic depth representation, and taking the interactive fusion result as the semantic correlation between the target bullet screen and the bullet screen segment.

[0123] The semantic feature of the target bullet screen is the bullet screen semantic depth representation, and the semantic feature of the bullet screen segment is the bullet screen segment semantic depth representation.

[0124] As shown in Figure 5 FIG. 1 is a schematic diagram of a process for obtaining semantic relevance in an embodiment. Referring to Figure 5 , the computer device can respectively pass the text of the target barrage and the text of the barrage segment through the BERT layer to obtain a barrage semantic deep representation representing the semantic features of the target barrage and a barrage segment semantic deep representation representing the semantic features of the barrage segment. Then, the barrage semantic deep representation and the barrage segment semantic deep representation are interactively fused through an attention mechanism layer, and the interactive fusion result is taken as the semantic relevance of the target barrage and the barrage segment.

[0125] In this embodiment, the barrage semantic deep representation of the target barrage and the barrage segment semantic deep representation of the barrage segment are interactively fused to ensure the accuracy of the semantic relevance of the target barrage and the barrage segment.

[0126] In some embodiments, the target barrage that is synchronously displayed with the target video segment is obtained by: obtaining a plurality of barrages that are synchronously displayed with the target video segment; performing scenario detection on each of the plurality of barrages respectively to obtain a scenario detection result representing whether the barrage has a scenario; and determining, according to the scenario detection result, a barrage having a scenario from the plurality of barrages as the target barrage.

[0127] Specifically, after obtaining the plurality of barrages that are synchronously displayed with the target video segment, the computer device can perform scenario detection on each of the barrages to determine whether the barrage has a scenario. If the scenario detection result represents that the barrage has a scenario, the barrage is taken as the target barrage for subsequent spoiler barrage identification.

[0128] In this embodiment, the detection of whether the barrage has a scenario is performed before the spoiler barrage identification, which improves the identification efficiency of the spoiler barrage.

[0129] In some embodiments, the scenario detection is performed on each of the plurality of barrages respectively, which can specifically include the following steps: performing word segmentation processing on each of the barrages to obtain a word segmentation sequence corresponding to each of the barrages; obtaining a scenario detection feature corresponding to the barrage according to each word in the word segmentation sequence and the position of each word in the word segmentation sequence; and inputting the scenario detection feature into a pre-trained scenario detection neural network to output a scenario detection result representing whether the barrage has a scenario through the scenario detection neural network.

[0130] The word segmentation of the barrage can obtain each barrage word included in the barrage and the position of each barrage word in the barrage. According to the position of each barrage word in the barrage, a word segmentation sequence formed by the barrage words is obtained. As shown in Figure 6 FIG. 2 is a schematic diagram of a process for detecting whether a barrage has a scenario in an embodiment. Referring to Figure 6The positions of the barrage words 0, 1, 2, and m in the barrage are positions 0, 1, 2, and m respectively. The scenario feature corresponding to the barrage includes the barrage words in the barrage and the positions of the barrage words in the barrage (i.e., the positions of each word in the word sequence).

[0131] The scenario detection neural network is a deep neural network for detecting whether a barrage has a scenario. The scenario detection neural network is obtained by training a model on barrage data with scenarios and barrage data without scenarios that are constructed in advance.

[0132] The computer device inputs the barrage content into the scenario detection neural network, and the scenario detection neural network outputs a probability that the barrage has a scenario as a scenario detection result. When the probability that the barrage has a scenario is greater than or equal to a preset threshold, it is determined that the barrage has a scenario, and at this time, the barrage can be determined to be a spoiler barrage.

[0133] In this embodiment, the scenario detection neural network obtains a detection result of whether the barrage has a scenario according to the barrage words included in the barrage and the positions of the barrage words in the barrage, thereby improving the detection accuracy of whether the barrage has a scenario.

[0134] As shown in FIG. 1, it is a flowchart of a barrage identification method in an embodiment. The method can be executed by a computer device. Referring to FIG. 1, the method includes the following steps. Figure 7 Figure 7 The method includes the following steps.

[0135] In step S701, it is determined whether a target barrage displayed synchronously with a target video segment has a scenario.

[0136] In step S702, it is determined whether the target barrage is a spoiler barrage for a subsequent video segment.

[0137] In step S703, when the target barrage is a spoiler barrage for a subsequent video segment, the target barrage is processed based on a viewing situation of the user for the subsequent video segment.

[0138] In an embodiment, before determining whether the target barrage is a spoiler barrage, the computer device can first determine whether the target barrage has a scenario. The scenario determination can be performed by the flow as shown in FIG. 2. The BERT layer determines whether the barrage has a scenario according to the barrage words in the barrage and the positions of the barrage words in the barrage. Figure 6

[0139] As shown in FIG. 3, it is a flowchart of a barrage identification method in an embodiment. The method can be executed by a computer device. Referring to FIG. 3, the method includes the following steps. Figure 8 Figure 8 The method includes the following steps.

[0140] ​​​Step S801, obtain a plurality of bullet screens displayed synchronously with a target video segment;

[0141] Step S802, perform word segmentation processing on each bullet screen to obtain a word segmentation sequence corresponding to each bullet screen;

[0142] Step S803, obtain a plot detection feature corresponding to each bullet screen according to each word in the word segmentation sequence and the position of each word in the word segmentation sequence;

[0143] Step S804, input the plot detection feature into a pre-trained plot detection neural network, and output a plot detection result of whether the bullet screen has a plot through the plot detection neural network;

[0144] Step S805, according to the plot detection result, take the bullet screen with a plot in the plurality of bullet screens as a target bullet screen.

[0145] Step S806, determine a subsequent video segment whose playback sequence is after the target video segment;

[0146] Step S807, obtain a bullet screen plot depth representation of the target bullet screen, and obtain a video segment plot depth representation of the subsequent video segment;

[0147] Step S808, interactively fuse the bullet screen plot depth representation and the video segment plot depth representation, and take the interactive fusion result as a relevance of the target bullet screen and the subsequent video segment in a plot;

[0148] Step S809, obtain a bullet screen segment formed by bullet screens displayed synchronously with the subsequent video segment, obtain a bullet screen semantic depth representation of the target bullet screen, and obtain a bullet screen segment semantic depth representation of the bullet screen segment;

[0149] Step S810, interactively fuse the bullet screen semantic depth representation and the bullet screen segment semantic depth representation, and take the interactive fusion result as a relevance of the target bullet screen and the bullet screen segment in semantics;

[0150] Step S811, determine a total number of bullet screens published by a target account and having a publishing time before a publishing time of the target bullet screen, and determine a number of spoiler bullet screens published by the target account and having a publishing time before the publishing time of the target bullet screen;

[0151] Step S812, take a ratio of the number of spoiler bullet screens to the total number as a tendency degree of the target account in publishing spoiler bullet screens.

[0152] Step S813, obtain a first weight corresponding to the relevance in a plot and the relevance in semantics, and a second weight corresponding to the tendency degree;

[0153] Step S814, obtaining the probability that the target barrage is a spoiler barrage according to the product result between the first weight, the relevance in plot and the relevance in semantics, and the product result between the second weight and the tendency.

[0154] The above barrage identification method, when identifying whether the target barrage is a spoiler barrage, after obtaining the target barrage and a subsequent video segment that are synchronously displayed with a target video segment and whose playing order is after the target video segment, obtains the relevance in plot between the target barrage and the subsequent video segment based on comparison between the plot feature of the target barrage and the plot feature of the subsequent video segment, obtains the relevance in semantics between the target barrage and a barrage segment synchronously displayed with the subsequent video based on comparison between the semantic feature of the target barrage and the semantic feature of the barrage segment synchronously displayed with the subsequent video, and then identifies the target barrage based on the relevance in plot and the relevance in semantics to obtain the identification result of whether the target barrage is a spoiler barrage. On the one hand, based on the definition of the spoiler barrage, the identification of the spoiler barrage depends not only on the target barrage itself but also on the relevance between the target barrage and the candidate video segment, and on the other hand, the relevance includes two dimensions of the plot relevance and the semantic relevance, which fully considers the internal relationship between the target barrage and the candidate video segment, thereby greatly improving the accuracy of identifying the spoiler barrage.

[0155] It should be understood that, although Figures 2 to 8 the steps in the flowchart of the method are shown in sequence according to the arrows, these steps are not necessarily executed in sequence according to the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, Figures 2 to 8 at least part of the steps in the method can include multiple steps or multiple stages, which are not necessarily executed at the same time but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps. It should be noted that the above description of "first", "second", etc. is used to distinguish similar objects and does not necessarily describe a specific sequence or order.

[0156] In some embodiments, as shown in Figure 9 a barrage identification device is provided, comprising:

[0157] The target barrage acquisition module 901 is configured to acquire a target barrage synchronously displayed with a target video segment.

[0158] The subsequent video segment acquisition module 902 is configured to determine a subsequent video segment whose playing order is after the target video segment.

[0159] The plot relevance obtaining module 903 is configured to take the comparison result between the plot feature of the target barrage and the plot feature of the subsequent video segment as the plot relevance between the target barrage and the subsequent video segment.

[0160] The semantic relevance obtaining module 904 is configured to obtain a barrage segment formed by the barrage displayed synchronously with the subsequent video segment, and take the comparison result between the semantic feature of the target barrage and the semantic feature of the barrage segment as the semantic relevance between the target barrage and the barrage segment.

[0161] The spoiler barrage identification module 905 is configured to identify the target barrage based on the plot relevance and the semantic relevance, and obtain the identification result of whether the target barrage is a spoiler barrage.

[0162] In some embodiments, the spoiler barrage identification module 905 is further configured to obtain a tendency degree of a target account in publishing spoiler barrages, the target account being an account that publishes the target barrage; obtain a probability that the target barrage is a spoiler barrage based on a positive correlation between each of the plot relevance, the semantic relevance, and the tendency degree and a barrage spoiler probability; and obtain the identification result of whether the target barrage is a spoiler barrage based on the probability that the target barrage is a spoiler barrage.

[0163] In some embodiments, the spoiler barrage identification module 905 is further configured to obtain a first weight corresponding to the plot relevance and the semantic relevance, and a second weight corresponding to the tendency degree; and obtain the probability that the target barrage is a spoiler barrage according to a product result between the first weight, the plot relevance, and the semantic relevance, and a product result between the second weight and the tendency degree.

[0164] In some embodiments, the spoiler barrage identification module 905 is further configured to determine a publishing time of the target barrage; determine a total number of barrages published by the target account and having a publishing time before the publishing time of the target barrage; determine a number of spoiler barrages published by the target account and having a publishing time before the publishing time of the target barrage; and take a ratio of the number of spoiler barrages to the total number as the tendency degree of the target account in publishing spoiler barrages.

[0165] In some embodiments, the plot feature of the target barrage is the barrage plot depth representation, and the plot feature of the subsequent video segment is the video segment plot depth representation; the plot relevance acquisition module 903 is also used to obtain the barrage plot depth representation of the target barrage and obtain the video segment plot depth representation of the subsequent video segment; interactively fuse the barrage plot depth representation and the video segment plot depth representation, and use the interactive fusion result as the plot relevance between the target barrage and the subsequent video segment.

[0166] In some embodiments, the plot relevance acquisition module is further configured to:

[0167] Performing speech recognition on the dialogue of the subsequent video segment to obtain a dialogue text of the subsequent video segment; constructing a video segment plot depth representation of the subsequent video segment based on the dialogue text;

[0168] And / or, performing character recognition on the video frames of the subsequent video segment to obtain subtitle text of the subsequent video segment; and constructing a video segment plot depth representation of the subsequent video segment based on the subtitle text.

[0169] In some embodiments, the semantic feature of the target barrage is the barrage semantic depth representation, and the semantic feature of the barrage segment is the barrage segment semantic depth representation; the semantic relevance acquisition module 904 is also used to obtain the barrage semantic depth representation of the target barrage and obtain the barrage segment semantic depth representation of the barrage segment; interactively fuse the barrage semantic depth representation and the barrage segment semantic depth representation, and use the interactive fusion result as the semantic relevance between the target barrage and the barrage segment.

[0170] In some embodiments, the target barrage acquisition module is also used to obtain multiple barrages displayed synchronously with the target video clip; perform plot detection on each of the multiple barrages to obtain a plot detection result that characterizes whether the barrage has a plot; and determine the barrage with a plot among the multiple barrages according to the plot detection result as the target barrage.

[0171] In some embodiments, the target barrage acquisition module 901 is further used to perform word segmentation processing on each barrage to obtain a word segmentation sequence corresponding to each barrage; obtain the plot detection feature corresponding to the barrage based on each word in the word segmentation sequence and the position of each word in the word segmentation sequence; input the plot detection feature into a pre-trained plot detection neural network, and output a plot detection result of whether the barrage has a plot through the plot detection neural network.

[0172] In some embodiments, the plot relevance acquisition module 903 is further used to input the plot features of the target barrage and the plot features of the subsequent video segment into a pre-trained plot relevance detection neural network for comparison; and use the comparison result output by the plot relevance detection neural network as the plot relevance between the target barrage and the subsequent video segment.

[0173] In some embodiments, the device also includes a plot relevance detection neural network training module, which is used to obtain multiple sample barrages displayed synchronously with the sample video clip; determine the first sample barrage and the second sample barrage among the multiple sample barrages; when the audience recognition corresponding to the first sample barrage is higher than the audience recognition corresponding to the second sample barrage, determine that the plot relevance of the first sample barrage to the sample video clip is higher than the plot relevance of the second sample barrage to the sample video clip; based on the fact that the plot relevance of the first sample barrage to the sample video clip is higher than the plot relevance of the second sample barrage to the sample video clip, the plot features of the first sample barrage, the plot features of the second sample barrage and the plot features of the sample video clip, train the plot relevance detection neural network.

[0174] In some embodiments, the device also includes an audience recognition determination module, which is used to obtain the number of likes of the first sample barrage and the number of likes of the second sample barrage; when the number of likes of the first sample barrage is greater than the number of likes of the second sample barrage, it is determined that the audience recognition corresponding to the first sample barrage is higher than the audience recognition corresponding to the second sample barrage.

[0175] In the above-mentioned barrage recognition device, when it is necessary to identify a target barrage as a spoiler barrage, after obtaining the target barrage displayed synchronously with the target video segment and the subsequent video segment whose playback order follows the target video segment, the plot relevance between the target barrage and the subsequent video segment is obtained based on the comparison between the plot features of the target barrage and the plot features of the subsequent video segment; the semantic relevance between the target barrage and the barrage segment displayed synchronously with the subsequent video segment is obtained based on the comparison between the semantic features of the target barrage and the semantic features of the barrage segment displayed synchronously with the subsequent video segment; then, based on the plot relevance and the semantic relevance, the target barrage is identified to obtain an identification result of whether the target barrage is a spoiler barrage. On the one hand, based on the definition of spoiler barrage, the identification of spoiler barrage does not rely solely on the target barrage itself, but relies on the relevance between the target barrage and the candidate video segment to identify spoiler barrage. On the other hand, the relevance includes two dimensions: plot relevance and semantic relevance, fully considering the intrinsic connection between the target barrage and the candidate video segment, thereby greatly improving the accuracy of identifying spoiler barrage.

[0176] For the specific definition of the barrage recognition device, please refer to the definition of the barrage recognition method above, which will not be repeated here. The various modules in the above-mentioned barrage recognition device can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0177] In some embodiments, a computer device is provided, which may be Figure 1 The terminal or server shown in FIG. 1 may have an internal structure as shown in FIG. Figure 10 As shown. The computer device includes a processor, a memory and a communication interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store barrage data. The network interface of the computer device is used to communicate with an external computer device via a network connection. When the computer program is executed by the processor, a barrage recognition method is implemented.

[0178] Those skilled in the art will understand that Figure 10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0179] In some embodiments, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned various method embodiments when executing the computer program.

[0180] In some embodiments, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned various method embodiments are implemented.

[0181] In some embodiments, a computer program product is provided, including a computer program, which implements the steps in the above-mentioned various method embodiments when executed by a processor.

[0182] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0183] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the above-mentioned computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0184] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0185] The above embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for identifying barrage, characterized in that: The method comprises: Get the target barrage displayed synchronously with the target video clip; Determine the subsequent video segments that follow the target video segment in a playback order; The comparison result between the plot features of the target bullet comment and the plot features of the subsequent video segment is used as the relevance of the target bullet comment and the subsequent video segment in terms of the plot; Obtaining a bullet comment segment formed by a bullet comment displayed synchronously with the subsequent video segment, and using a comparison result between semantic features of the target bullet comment and semantic features of the bullet comment segment as the semantic relevance between the target bullet comment and the bullet comment segment; Based on the relevance in plot and the relevance in semantics, the target barrage is identified to obtain an identification result of whether the target barrage is a spoiler barrage.

2. The method according to claim 1, characterized in that Based on the plot relevance and the semantic relevance, the target barrage is identified to obtain an identification result of whether the target barrage is a spoiler barrage, including: Obtaining a target account's propensity to post spoiler comments; the target account is the account that posted the target comment; Based on the positive correlation between the plot relevance, the semantic relevance, and the tendency and the spoiler probability of the bullet screen, a probability that the target bullet screen is a spoiler bullet screen is obtained; Based on the probability that the target barrage is a spoiler barrage, an identification result of whether the target barrage is a spoiler barrage is obtained.

3. The method according to claim 2, characterized in that Based on the positive correlation between the plot relevance, the semantic relevance, and the tendency and the spoiler probability of the bullet screen, the probability that the target bullet screen is a spoiler bullet screen is obtained, including: Obtaining a first weight corresponding to the plot relevance and the semantic relevance, and a second weight corresponding to the tendency; The probability that the target barrage is a spoiler barrage is obtained based on the product of the first weight, the plot relevance and the semantic relevance, and the product of the second weight and the tendency.

4. The method according to claim 2, characterized in that Obtaining the target account's propensity to post spoiler comments includes: Determining the release time of the target barrage; Determine the total number of bullet comments posted by the target account and published before the target bullet comment, and determine the number of spoiler bullet comments posted by the target account and published before the target bullet comment; The ratio of the number of spoiler bullet comments to the total number is used as the tendency of the target account to post spoiler bullet comments.

5. The method according to claim 1, wherein The plot feature of the target barrage is the barrage plot depth representation, and the plot feature of the subsequent video segment is the video segment plot depth representation; The comparing result between the plot features of the target bullet comment and the plot features of the subsequent video segment as the relevance of the target bullet comment and the subsequent video segment in terms of the plot includes: Obtaining a bullet comment plot depth representation of the target bullet comment, and obtaining a video clip plot depth representation of the subsequent video clip; The plot depth representation of the barrage and the plot depth representation of the video clip are interactively fused, and the interactive fusion result is used as the plot relevance between the target barrage and the subsequent video clip.

6. The method according to claim 5, characterized in that Obtaining a video segment plot depth representation of the subsequent video segment, including: Performing speech recognition on the dialogue of the subsequent video segment to obtain the dialogue text of the subsequent video segment; constructing a video segment plot depth representation of the subsequent video segment based on the dialogue text; and / or, Character recognition is performed on the video frames of the subsequent video segment to obtain subtitle text of the subsequent video segment; and based on the subtitle text, a video segment plot depth representation of the subsequent video segment is constructed.

7. The method according to claim 1, characterized in that The semantic feature of the target barrage is the barrage semantic depth representation, and the semantic feature of the barrage segment is the barrage segment semantic depth representation; The comparing result between the semantic features of the target barrage and the semantic features of the barrage segment as the semantic relevance between the target barrage and the barrage segment includes: Obtaining a barrage semantic depth representation of the target barrage, and obtaining a barrage segment semantic depth representation of the barrage segment; The semantic depth representation of the barrage and the semantic depth representation of the barrage segment are interactively fused, and the interactive fusion result is used as the semantic relevance between the target barrage and the barrage segment.

8. The method according to claim 1, characterized in that The step of obtaining a target bullet comment displayed synchronously with the target video clip includes: Get multiple comments that are displayed synchronously with the target video clip; Performing plot detection on each of the plurality of barrages to obtain a plot detection result indicating whether the barrage has a plot; According to the plot detection result, a barrage with a plot among the multiple barrages is determined as a target barrage.

9. The method according to claim 8, characterized in that The performing plot detection on each of the plurality of bullet comments separately includes: Perform word segmentation on each barrage to obtain the corresponding word segmentation sequence for each barrage; Obtaining plot detection features corresponding to the barrage according to each word in the word segmentation sequence and the position of each word in the word segmentation sequence; The plot detection features are input into a pre-trained plot detection neural network, and the plot detection neural network outputs a plot detection result of whether the barrage has a plot.

10. The method according to claim 1, characterized in that The comparing result between the plot features of the target bullet comment and the plot features of the subsequent video segment as the relevance of the target bullet comment and the subsequent video segment in terms of the plot includes: Input the plot features of the target barrage and the plot features of the subsequent video clips into a pre-trained plot relevance detection neural network for comparison; The comparison result output by the plot relevance detection neural network is used as the plot relevance between the target barrage and the subsequent video clip.

11. The method according to claim 10, characterized in that The method further comprises: Obtain multiple sample comments that are displayed synchronously with the sample video clips; Determining a first sample barrage and a second sample barrage from the plurality of sample barrages; When the audience recognition corresponding to the first sample barrage is higher than the audience recognition corresponding to the second sample barrage, determining that the relevance of the first sample barrage to the sample video clip in terms of plot is higher than the relevance of the second sample barrage to the sample video clip in terms of plot; Based on the fact that the plot relevance between the first sample barrage and the sample video clip is higher than the plot relevance between the second sample barrage and the sample video clip, the plot features of the first sample barrage, the plot features of the second sample barrage and the plot features of the sample video clip, the plot relevance detection neural network is trained.

12. The method according to claim 11, characterized in that The method further comprises: Obtain the number of likes for the first sample barrage and the number of likes for the second sample barrage; When the number of likes for the first sample barrage is greater than the number of likes for the second sample barrage, it is determined that the audience recognition corresponding to the first sample barrage is higher than the audience recognition corresponding to the second sample barrage.

13. A barrage recognition device, characterized in that: The device comprises: Target barrage acquisition module, used to obtain the target barrage displayed synchronously with the target video clip; A subsequent video segment acquisition module, configured to determine subsequent video segments that follow the target video segment in a playback order; a plot relevance acquisition module, configured to use a comparison result between the plot features of the target barrage and the plot features of the subsequent video segment as the plot relevance between the target barrage and the subsequent video segment; a semantic relevance acquisition module, configured to acquire a bullet comment segment formed by a bullet comment displayed synchronously with the subsequent video segment, and use a comparison result between the semantic features of the target bullet comment and the semantic features of the bullet comment segment as the semantic relevance between the target bullet comment and the bullet comment segment; The spoiler barrage recognition module is used to identify the target barrage based on the relevance in the plot and the relevance in the semantics, and obtain an identification result of whether the target barrage is a spoiler barrage.

14. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 12 is implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

16. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

Images