Topic generation method and device
By performing semantic recognition on the target comment content of multimedia data, discussion topics are generated, solving the problem of low efficiency in traditional topic generation and achieving more efficient and objective discussion topic generation and improved user engagement.
Patent Information
- Application Number
- CN202110786949.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-07-12
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2041-07-12
AI Technical Summary
Traditional topic generation is inefficient, requires a lot of human resources, and cannot effectively improve user engagement on multimedia data platforms.
By performing semantic recognition on the target comment content of multimedia data, discussion topics are generated, avoiding human editing. Natural language processing technology is used to identify semantic information and generate discussion topics.
It improves the efficiency of topic generation, makes the generated discussion topics more objective, can continuously attract attention, increases the amount of topics in multimedia data, and saves human resources.
Smart Images

Figure CN113822077B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a topic generation method and device. Background Art
[0002] In order to increase user participation, most multimedia data platforms will set up a topic discussion area, which can include at least one hot topic. However, traditional topics are manually edited, which requires a lot of human resources, resulting in low topic generation efficiency. Therefore, how to improve the efficiency of topic generation has become a current research hotspot. Summary of the invention
[0003] The embodiments of the present application provide a topic generation method and device, which can improve the efficiency of topic generation.
[0004] On the one hand, an embodiment of the present application provides a topic generation method, including:
[0005] Performing semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information; wherein the topic degree corresponding to the target comment content or the target semantic information satisfies a preset topic condition, and the topic degree is used to indicate the degree of participation in the multimedia data;
[0006] Based on the target semantic information, a discussion topic of the multimedia data is generated.
[0007] Accordingly, an embodiment of the present application provides a topic generation device, including:
[0008] A semantic recognition unit, configured to perform semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information; wherein the topic degree corresponding to the target comment content or the target semantic information satisfies a preset topic condition, and the topic degree is used to indicate the degree of participation in the multimedia data;
[0009] A generating unit is used to generate a discussion topic of the multimedia data based on the target semantic information.
[0010] Accordingly, the present application provides a topic generation device, comprising:
[0011] a processor adapted to implement one or more instructions;
[0012] A computer storage medium storing one or more instructions, wherein the one or more instructions are suitable for being loaded by the processor and executing the above topic generating method.
[0013] Accordingly, an embodiment of the present application provides a computer storage medium, characterized in that the computer storage medium stores one or more instructions, and the one or more instructions are suitable for being loaded by the processor and executing the above-mentioned topic generation method.
[0014] Accordingly, an embodiment of the present application provides a computer program product or a computer program, which includes a computer program, and the computer program is stored in a computer storage medium; the processor of the topic generation device reads the computer program from the computer storage medium, and the processor executes the computer program, so that the topic generation device performs the above-mentioned topic generation method.
[0015] The topic generation method provided in the embodiment of the present application performs semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information, and then generates discussion topics of multimedia data based on the target semantic information. Since the discussion topics are generated after semantic understanding of one or more target comment contents, the subjectivity of manually editing the discussion topics is avoided, and the efficiency of topic generation can be improved to a certain extent. In addition, since exclusive discussion topics are generated for discussion hotspots, the multimedia data can have more topic volume. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0017] Figure 1 It is a flowchart of a topic generation method provided in an embodiment of the present application;
[0018] Figure 2 This is a schematic diagram of a bullet screen content provided by an embodiment of the present application;
[0019] Figure 3 It is a flowchart of generating discussion topics based on bullet screen content provided by an embodiment of the present application;
[0020] Figure 4 is a schematic flow chart of a method for determining a target multimedia segment based on a discussion topic provided by an embodiment of the present application;
[0021] Figure 5 is a schematic flow chart of another method for determining a target multimedia segment based on a discussion topic provided in an embodiment of the present application;
[0022] Figure 6aIt is a schematic diagram of a face recognition process provided by an embodiment of the present application;
[0023] Figure 6b It is a flow chart of an action recognition process provided by an embodiment of the present application;
[0024] Figure 7a It is a schematic diagram of a process of determining a target multimedia segment matching a discussion topic provided by an embodiment of the present application;
[0025] Figure 7b This is another flowchart of determining a target multimedia segment matching a discussion topic provided by an embodiment of the present application;
[0026] Figure 7c This is another flowchart of determining a target multimedia segment matching a discussion topic provided by an embodiment of the present application;
[0027] Figure 7d It is a flowchart of generating discussion topics based on the comments in the comment area provided by an embodiment of the present application;
[0028] Figure 8 is a schematic diagram of a method for triggering display of a target multimedia segment provided by an embodiment of the present application;
[0029] Fig. 9 It is a structural schematic diagram of a topic generating device provided in an embodiment of the present application;
[0030] Fig.10 It is a structural diagram of a topic generating device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0031] The present application provides a topic generation method, which can automatically generate discussion topics for multimedia data, especially for multimedia data with high topicality, so as to increase the topicality of the multimedia data, and the topicality is used to indicate the degree of participation in the multimedia data. Among them, multimedia data with high topicality can refer to: multimedia data that has aroused widespread attention and heated discussion among the masses in the media, and the multimedia data can be video data (such as: videos of well-known stars participating in programs), text data (such as: text reports of major natural disasters), and audio data (such as: radio publicity of important activities); discussion topics can refer to: speech themes (or: topic points) obtained after a concise summary of various comments made by the masses on the multimedia data. For example, suppose video A is a video of little a performing a dance for children at a public welfare activity for building a school. After watching video A, a large number of netizens posted comments in the comment section of video A. In this case, if a large number of comments in the comment section are praising little a for dancing well, then the discussion topic for the multimedia data in the comment section (i.e., video A) can be: little a dance is great; if a large number of comments in the comment section are calling on more people to learn from little a and participate in public welfare activities, then the discussion topic for the multimedia data in the comment section can be: participate in public welfare activities with little a.
[0032] The following is a description of the general principle of the topic generation method provided in the embodiment of the present application. The topic generation method mainly performs semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information, and further, generates discussion topics based on the target semantic information. Among them, the one or more target comment contents mentioned above can specifically refer to multiple comment contents. In this case, the target semantic information can be understood as: the semantics expressed by most of the comment contents in the multiple comment contents. Further, it can be stipulated that the topicality of the target semantic information meets the preset topic conditions; optionally, the one or more target comment contents mentioned above can also specifically refer to one or more specific comment contents, and the topicality of each comment content in the above-mentioned one or more specific comment contents meets the preset topic conditions, such as comment contents with more likes or more views, which can be understood as comment contents with a higher degree of participation in the multimedia data, that is, comment contents with a topicality that meets the preset topic conditions. In this case, the target semantic information can be understood as the semantics expressed by the above-mentioned one or more specific comment contents, and a discussion topic can be generated based on the semantic information of each comment content. It should be noted that the preset topic conditions mentioned above may include but are not limited to any of the following: the topic is ranked Nth (such as the 1st, 2nd, and 3rd) after being arranged from high to low, the topic meets the topic threshold, the topic is ranked in the top N (such as the top 2 and top 3) after being arranged from high to low, etc., where N is a positive integer and is less than or equal to the number of one or more comments.
[0033] Based on the above description, it can be understood that since the topic generation method provided in the embodiment of the present application generates discussion topics based on the semantic information of the target comment content after semantic recognition of the target comment content of the multimedia data, the embodiment of the present application can avoid manual participation in the generation of discussion topics (such as: avoiding manual editing of discussion topics), realizes the automatic generation of discussion topics, and can save human resources to a certain extent.
[0034] In one embodiment, the topic generation method can be executed by a topic generation device, and the topic generation device can be a server. Specifically, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms, etc. The embodiments of the present application do not limit this.
[0035] See also Figure 1 , Figure 1is a flowchart of a topic generation method provided in an embodiment of the present application. The topic generation method can be executed by the topic generation device mentioned above, such as Figure 1 As shown, the method includes steps S101-S102:
[0036] S101, performing semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information.
[0037] The target comment content may be the bullet content corresponding to the multimedia data. The bullet content refers to the commentary subtitles that pop up when the multimedia data is watched online via the Internet. The bullet content may be displayed superimposed on the playback interface of the multimedia data, and the duration of the superimposed display of the bullet content on the playback interface of the multimedia data is a preset duration. The preset duration is generally not longer than the total playback duration of the multimedia data. For example, when the total playback duration of the multimedia data is 20 seconds, the preset duration may be, for example, 3 seconds, 5 seconds, etc. Figure 2 As shown, the bullet content 201 stops being displayed on the multimedia data playback interface after a preset time. It should be noted that as the multimedia data is played, the bullet content displayed on the terminal device may change, and the changes that occur include but are not limited to any one or more of the following: changes in display position (such as: moving to the left, moving to the right, etc.), changes in quantity (such as: increasing quantity, decreasing quantity, etc.), see Figure 2 As shown, both the bullet content 201 and the bullet content 202 are superimposed and displayed on the playback interface of the multimedia data. After a period of time (e.g., a preset duration), the bullet content 202 stops being superimposed and displayed on the playback interface of the multimedia data, and the amount of bullet content displayed on the terminal device changes. The terminal device is a device that plays multimedia data, that is, a device that displays the playback interface of multimedia data. Optionally, the target comment content may also be the comment content in the comment area of the multimedia data, wherein the comment content in the comment area may include, but is not limited to, any one or more of the following: picture comment content, video comment content, text comment content, and audio comment content.
[0038] In specific applications, the topic generation device can use a natural language recognition model to perform semantic recognition on the comment content. It should be noted that the natural language recognition model is built based on natural language processing technology; natural language processing (NLP) technology is a technology that studies how computers process human natural language (i.e., the language people use in daily life). It is also an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. In addition, natural language processing is a science that integrates linguistics, computer science, and mathematics. Natural language processing technology usually includes text processing, semantic understanding (or semantic recognition), machine translation, robot question answering, knowledge graphs, and other technologies. Based on this, it can be understood that natural language processing can include natural language recognition, which can include: syntactic semantic analysis, information extraction, text mining, machine translation, information retrieval, etc. Then, it can be understood that the use of a natural language recognition model can perform semantic recognition on the target comment content, and then the semantic information of the target comment content can be obtained.
[0039] In one embodiment, as can be seen from the foregoing, one or more target comment contents may refer to multiple comment contents. In this case, one or more comment contents may be all the comment contents of the multimedia data (such as: all the comment contents in the comment area of the multimedia data, or all the barrage contents of the multimedia data), or may be multiple comment contents obtained after removing irrelevant comment contents from all the comment contents of the multimedia data. For example, assuming that there are 50 comment contents in the comment area of food video A, one of which is a comment content a for a certain cosmetics, then it can be considered that comment content a is a comment content irrelevant to food video A, and therefore the corresponding one or more target comment contents of food video A are the 49 comment contents after removing comment content a. Based on this, when the topic generation device performs semantic recognition on one or more target comment contents of multimedia data and obtains target semantic information, it can specifically include the following steps: the topic generation device first obtains semantic recognition on one or more target comment contents of multimedia data to obtain one or more semantic information. After the topic generation device obtains the one or more semantic information, it can obtain the number of comment contents corresponding to similar semantic information in the one or more semantic information; further, if there is a number of comment contents corresponding to target similar semantic information that meets the first preset number condition, it is determined that the topic degree corresponding to the target similar semantic information meets the preset topic condition, and then the topic generation device can use the target similar semantic information as the target semantic information. Among them, similar semantic information can be understood as: one or more similar semantic information, and when the similar semantic information includes multiple semantic information, the similarity between any two semantic information is greater than the similarity threshold. The target similar semantic information can be understood as: similar semantic information whose corresponding comment contents meet the first preset number condition. The first preset quantity condition may include but is not limited to any one of the following: the largest quantity, the quantity meeting a preset quantity threshold, the quantity being ranked in the top P positions after being sorted from most to least, etc., where P is a positive integer, and P is greater than zero and less than or equal to the quantity of similar semantic information.
[0040] Among them, the topic generation device obtains the number of comment contents corresponding to similar semantic information in one or more semantic information, which can be specifically implemented in the following manner: the topic generation device can determine one or more semantic information groups from one or more semantic information, the similarity between the semantic information contained in each semantic information group is greater than the preset ratio threshold, and each semantic information group can indicate a similar semantic information; then further, the topic generation device can obtain the sum of the number of comment contents corresponding to all semantic information in each semantic information group, and use the sum of the number of comment contents corresponding to the semantic information group as the number of comment contents corresponding to the similar semantic information indicated by the semantic information group. Since one or more semantic information can determine one or more semantic information groups, and each semantic information group indicates a similar semantic information, it is not difficult to understand that the number of similar semantic information is the number of semantic information groups (i.e.: one or more). Based on this, the target similar semantic information can be understood as: the similar semantic information indicated by the semantic information group whose sum meets the above-mentioned first preset quantity condition.
[0041] Optionally, the topic generation device may perform semantic recognition on one or more target comment contents of multimedia data to obtain one or more semantic information. After the topic generation device obtains the one or more semantic information, the number of comment contents corresponding to the same semantic information in the one or more semantic information may be obtained; further, if the number of comment contents corresponding to the semantic information satisfies the first preset quantity condition, it is determined that the topic degree corresponding to the semantic information satisfies the preset topic condition, and then the topic generation device may use the semantic information as the target semantic information. The first preset quantity condition may include but is not limited to any one of the following: the largest number, the number meets the preset quantity threshold, the number is ranked in the top M positions after sorting by number from most to least, etc., M is a positive integer, M is greater than zero and less than or equal to the number of semantic information.
[0042] In order to facilitate a clear understanding of the topic generation method provided by the present application, the following takes the barrage content as the target comment content as an example, and describes in detail the specific method for the topic generation device to obtain the target semantic information when one or more target comment contents are multiple comment contents: assuming that the first preset quantity condition is the maximum quantity, if there are a total of 100 barrage contents corresponding to the multimedia data A, after the topic generation device performs semantic understanding on each barrage content, it obtains 50 barrage contents with the semantic information "XX dances so handsome", 30 barrage contents with the semantic information "XX dances so standard", and 40 barrage contents with the semantic information "XX has a great body proportion". then, since both “XX dances so handsomely” and “XX dances to the standard” have the two keywords “XX” and “dance”, it can be considered that the similarity between the semantic information “XX dances so handsomely” and the semantic information “XX dances to the standard” is greater than the similarity threshold, and then the topic generating device can add the two semantic information “XX dances so handsomely” and “XX dances to the standard” to the same semantic information group (assuming it is semantic information group a), and add the semantic information “XX has a great body proportion” to another semantic information group (assuming it is semantic information group b). Furthermore, the topic device can obtain the number of barrage contents corresponding to each semantic information group. It is not difficult to see that the total number of barrage contents corresponding to the semantic information contained in semantic information group a is 80 (50+30=80), and the total number of barrage contents corresponding to the semantic information contained in semantic information group b is 20. It can be determined that the total number of comment contents corresponding to semantic information group a, 80, meets the first preset quantity condition. Therefore, the topic generation device can use the semantic information contained in semantic information group a as the target semantic information, that is, the target semantic information is: "XX dances so handsomely" and "XX dance moves are so standard". It should be noted that, similarly, when one or more target comment contents are multiple comment contents in the comment area, the topic generation device can also obtain the target semantic information in the manner described in the above example, which will not be elaborated in this application.
[0043] In another embodiment, as can be seen from the foregoing, one or more target comment contents may also refer to one comment content. In this case, one or more target comment contents may be one comment content whose topicality satisfies a preset topic condition among all the comment contents of the multimedia data. Among them, the topicality of any comment content among all the comment contents of the multimedia data can be obtained based on the interactive information of the any comment content, and the interactive information may include but is not limited to any of the following: the number of likes, the number of comments, and the number of views. Exemplarily, it can be stipulated that the level of topicality is positively correlated with the size of the interactive information, such as: it is stipulated that the more times a comment content is viewed, the higher its corresponding topicality; optionally, it can also be stipulated that the level of topicality is positively correlated with the growth rate of the interactive information within a preset time length, such as: within the preset time length, the more likes a comment content has, the higher its corresponding topicality. Based on this, when the topic generation device performs semantic recognition on one or more target comment contents of multimedia data and obtains target semantic information, it can specifically include the following steps: the topic generation device first obtains the interaction information of each comment content in the one or more comment contents of the multimedia data, and then determines the topicality corresponding to each comment content based on the interaction information of each comment content; further, the topic generation device can determine the target comment content whose topicality meets the preset topic conditions in one or more comment contents, and perform semantic recognition on the target comment content to obtain the semantic information of the target comment content, and then the topic generation device can use the semantic information of the target comment content as the target semantic information. Among them, the target comment content whose topicality meets the preset topic conditions may refer to: the comment content with the largest number of likes, comments or views, or the comment content ranked in the top L in terms of the number of likes, comments or views, or the comment content with the largest value obtained by performing an arithmetic operation (such as weighted sum or mean) on at least two of the number of likes, comments and views, or the comment content ranked in the top L in terms of the value obtained by performing an arithmetic operation on at least two of the number of likes, comments and views, where L is a positive integer, greater than zero and less than the number of one or more comment contents of the multimedia data.
[0044] Similarly, in order to facilitate a clear understanding of the topic generation method provided by the present application, the following takes the example that the topicality of a comment content is positively correlated with the number of likes for the comment content as an example, and describes in detail the specific method for the topic generation device to obtain the target semantic information when one or more target comment contents are a comment content: assuming that the preset topic condition is the highest topicality, if the multimedia data corresponds to 3,000 comment contents, then the topic generation device can use the comment content with the most likes among the 3,000 comment contents (assuming it is comment content a) as the comment content whose topicality meets the preset topic condition, and then obtain the target comment content; further, the topic device can perform semantic recognition on the comment content a to obtain the semantic information of the comment content a, then, the topic device can use the semantic information of the comment content a as the target semantic information.
[0045] It should be noted that, in actual applications, the target comment content can be any one of video comment content, audio comment content, picture comment content and text comment content, and this application does not limit this. When the target comment content is video comment content, audio comment content, or picture comment content, the topic generation device can obtain the semantic information of the target comment content in the following way: the topic generation device first performs content recognition on the target comment content to obtain the text information corresponding to the target comment content, and then performs semantic recognition on the text information to obtain the semantic information of the target comment content. Among them, content recognition can be understood as: recording the information expressed by the target comment content with text content, then the text information can refer to: text content that can be used to describe the semantics expressed by the target comment content. Exemplarily, the video comment content can be recognized by video understanding, the audio comment content can be recognized by voice recognition, and the picture comment content can be recognized by image recognition, and this application does not limit this. For example, suppose the target comment content is a video comment content, and the video comment content includes a video of dancer A dancing; if the video feature obtained by the topic generation device after understanding the video is: dancer A dances, and the audience applauds. Then the text information corresponding to the video comment content can be: text content describing the scene of "dancer A dancing", such as: "Dancer A is self-taught and his dancing is stunning." Based on the above description, it can be understood that when the target comment content is text comment content, the topic generation device can directly use the natural language recognition model to perform semantic recognition on the text comment content to obtain the semantic information of the text comment content.
[0046] S102: Generate discussion topics for multimedia data based on target semantic information.
[0047] Among them, the target comment content corresponding to the target semantic information can be part or all of the target comment content in one or more target comment contents. Based on this, the topic generation device can generate discussion topics in the following manner: the topic generation device performs word segmentation processing on all target comment contents corresponding to the target semantic information to obtain multiple keywords; wherein, each comment content can obtain at least one keyword after word segmentation processing, and different comment contents may obtain the same keyword after word segmentation processing; further, the topic generation device can perform semantic understanding on each of the multiple keywords, obtain keywords with similar semantic information, and determine the target keyword from the keywords with similar semantic information; further, the topic generation device can generate a discussion topic based on all the determined target keywords, and the discussion topic at least includes all the determined target keywords. Exemplarily, the discussion topic can include: all the determined target keywords, and other words (such as: adjectives, adverbs, prepositions, etc.) that modify all the determined target keywords.
[0048] In one embodiment, the target keyword can be the same keyword with the largest number among similar keywords, then the discussion topic at least includes all the keywords with the largest number mentioned above. The following is a detailed description of the steps for the topic generation device to generate discussion topics in this case in combination with a specific example: Assume that all target comment contents corresponding to the target semantic information are: comment content A "Xiao A sings so well", comment content B "I love listening to Xiao A singing, it's super healing", and comment content C "What kind of fairy voice is this, the singing voice is so healing". Then, the topic generating device may first perform word segmentation processing on the comment content A, comment content B and comment content C to obtain one or more keywords. If the topic generating device performs word segmentation processing on the comment content A "Xiao a sings so beautifully", the keywords obtained are "Xiao a", "singing", and "beautiful"; the topic generating device performs word segmentation processing on the comment content B "I like listening to Xiao a singing so much, it's super healing" and the keywords obtained are "Xiao a", "singing", and "healing"; the topic generating device performs word segmentation processing on the comment content C "What kind of fairy voice is this, the singing voice is so healing", and the keywords obtained are: "fairy voice", "singing voice", and "healing"; then, if the topic generating device performs word segmentation processing on the target language After word segmentation of all target comment contents corresponding to the semantic information, one or more groups of keywords with similar semantic information are obtained as follows: group a {"little a", "little a"}, group b {"singing", "singing", "singing voice"}, group c {"good-sounding", "healing", "healing", "divine voice"}; then the topic generation device can obtain the number of identical keywords in each group of similar keywords, and obtain: the number of "little a" in group a is 2; the number of "singing" in group b is 2, and the number of "singing voice" is 1; the number of "good-sounding" in group c is 1, the number of "healing" is 2, and the number of "divine voice" is 1; then it can be seen that the keywords with the largest number among all similar keywords include: "little a", "singing", and "healing". Therefore, the topic generation device can generate discussion topics based on "little a", "singing", and "healing". For example, the discussion topic can include only these three keywords, such as: little a singing heals; it can also include these three keywords and other words that describe the keywords (adjectives, verbs, prepositions, etc.), such as: everyone praises little a singing for healing people's hearts, and this application does not impose any restrictions on this.
[0049] In another embodiment, the target keyword may be the keyword with the lowest meaning among the keywords with similar semantic information. The lowest meaning may be understood as: the scope of the meaning is smaller and the things referred to are more specific; for example, the semantic information of the three keywords dance, street dance and breakdance are all "dance", but breakdance is a type of street dance, and street dance is a type of dance. Therefore, breakdance is the word with the lowest meaning among these three keywords with similar semantic information. Then, in specific applications, the topic generation device may take breakdance as the target keyword.
[0050] In one embodiment, based on the above description of steps S101 to S102, the specific manner in which the topic generating device generates the discussion topic of the multimedia data based on the bullet screen content of the multimedia data can be exemplified by referring to Figure 3 As shown, Figure 3 As shown, the topic generation device can first obtain all the barrage contents corresponding to the multimedia data, and then understand the obtained barrage contents (i.e., semantic recognition), and aggregate the barrage contents with similar semantic information, wherein content aggregation can be understood as: grouping the barrage contents with similar semantic information together; then, based on this, the topic generation device can determine the weight of the semantic information based on the aggregated content (i.e., the number of barrage contents with similar semantic information), then, further, the topic generation device can prioritize the semantic information according to the weight of the semantic information. For example, it can be specified that the barrage contents with similar semantic information The more barrage contents there are, the greater the weight of the similar semantic information is, wherein the weights of the various semantic information contained in the similar semantic information are the same. Based on this, the topic generation device can give priority to using the barrage contents with the largest semantic information weight to generate discussion topics; such as: if there are 5 barrage contents of similar semantic information a, 6 barrage contents of similar semantic information b, and 2 barrage contents of similar semantic information c, then the similar semantic information is arranged in descending order according to weight: similar semantic information b>similar semantic information a>similar semantic information c, then the topic generation device can generate discussion topics based on similar semantic information b.
[0051] The topic generation method provided in the embodiment of the present application performs semantic recognition on the target comment content of the multimedia data, and then uses the similar semantic information whose target comment content of similar semantic information meets the first preset quantity threshold as the target semantic information, and generates the discussion topic of the multimedia data based on the target semantic information. It can be seen that the topic generation method provided in the embodiment of the present application is generated based on the topic points with higher discussion volume. Compared with the traditional method of manually editing discussion topics, the topic generation method provided in the embodiment of the present application more accurately locates the discussion hotspots about the multimedia data, and the generated discussion topics are more objective, and can ensure that the discussion topics can continue to receive attention, thereby bringing more topic volume to the multimedia data corresponding to the discussion topic. It can be further understood that since the embodiment of the present application avoids manually editing discussion topics, it improves the efficiency of topic generation to a certain extent, while saving human resources.
[0052] Based on the relevant description of the above embodiment, the topic generation method provided in the embodiment of the present application can also further associate and store the generated discussion topic and the target multimedia segment matching the discussion topic after the topic generation device generates the discussion topic, so that the user can directly view the target multimedia segment related to the discussion topic on the topic discussion page, so that the user can more quickly understand the discussion content of the discussion topic, and to a certain extent improve the user experience when the user participates in the topic discussion. Among them, the target multimedia segment can be understood as: a multimedia segment strongly related to the discussion topic in the multimedia data, and the so-called strong correlation with the discussion topic can be understood as: there are features (such as pictures, sounds, text, etc.) indicated by the corresponding keywords of the discussion topic; for example, assuming that the discussion topic is "Little A dances beautifully", then the multimedia segment strongly related to "Little A dances beautifully" can be: a multimedia segment including a picture of Little A dancing in the multimedia data. In addition, it should be noted that the user can also selectively view other multimedia segments or the entire multimedia data in the multimedia data where the target multimedia segment is located in the process of viewing the target multimedia segment. In actual applications, the topic generation device mentioned above can be used to associate the discussion topic with the target multimedia segment, or other devices (such as: associating devices) can be used to associate the discussion topic with the target multimedia segment, wherein the associating device mentioned above can be a server, server cluster, etc., and the embodiments of the present application do not impose any restrictions on this.
[0053] See also Figure 4 , Figure 4 is a schematic flow chart of a method for determining a target multimedia segment based on a discussion topic provided by an embodiment of the present application, which can be applied to a scenario where one or more target comment contents are barrage contents; next, taking the topic generation device as an example to realize associating the discussion topic with the target multimedia segment, Figure 4 The method shown is described in detail; Figure 4 As shown, when the topic generation device associates the discussion topic with the target multimedia segment, the steps specifically include steps S401 to S406:
[0054] S401: Obtain the playback duration of multimedia data.
[0055] Among them, multimedia data may include but is not limited to any of the following: video data, audio data, text data, and picture data. When the multimedia data is video data or audio data, the playback duration may refer to: the duration of the video data or audio data being played on the device; when the multimedia data is text data or picture data, the playback duration may be understood as: the data size of the text data or picture data, or the total display duration of the text data or video data. Next, for the sake of ease of description, unless otherwise specified, this application takes the multimedia data as video data as an example to elaborate on the topic generation method provided by this application.
[0056] S402: Divide the playing time into time periods to obtain a plurality of playing time periods.
[0057] In a specific application, the topic generation device can evenly divide the playback duration into multiple playback time periods to obtain multiple playback time periods with consistent durations; optionally, the topic generation device can also unevenly divide the playback duration into multiple playback time periods according to actual needs to obtain multiple playback time periods with inconsistent durations. It should be noted that multiple playback time periods with inconsistent durations can refer to: among the multiple playback time periods, the durations of any two playback time periods are inconsistent; or can refer to: among the multiple playback time periods, there are any two playback time periods with inconsistent durations. This application does not impose any restrictions on this. Unless otherwise specified, this application uses the example of the consistency of the duration of each playback time period in the multiple playback time periods to elaborate on this application in detail.
[0058] S403: Obtain the number of segment comment contents of the multimedia segments played in each of the multiple play time periods.
[0059] In a specific implementation, the topic generation device may first obtain the time information of each target comment content in one or more target comment contents during the multimedia data playback process; and then determine the segment comment content of the multimedia segments played in each playback time period according to the time information of each target comment content and each playback time period. Further, the topic generation device may obtain the number of segment comment content of the multimedia segments played in each playback time period. Among them, the multimedia segments played in each playback time period are multimedia segments of the multimedia data; the segment comment content refers to: the comment content whose semantic information is the target semantic information, then it can be understood that the segment comment content is part or all of the comment content in one or more target comment contents corresponding to the multimedia data; the time information may include but is not limited to any of the following: display time, release time. It should be noted that the display time and release time mentioned here are both for the playback duration of the multimedia data, that is, the display time can be the time corresponding to the playback duration when the barrage content is displayed during the playback of the multimedia data; the release time can be a specific playback time point in the playback duration of the multimedia data.
[0060] The specific implementation of step S403 is described in detail below with reference to a specific example. Assuming that the playing time of multimedia data A is 20 minutes and the length of each playing time period is 5 minutes, the topic generating device can obtain 4 playing time periods after dividing the playing time of the multimedia data into time periods, namely: 0-5 minutes, 6-10 minutes, 11-15 minutes, and 16-20 minutes; based on this, if the time information is the release time, then when the multimedia data A corresponds to 10 barrage contents with semantic information as the target semantic information, of which 3 barrage contents are released at the time when multimedia data A is played to 4 minutes and 30 seconds, 6 barrage contents are released at the time when multimedia data is played to 6 minutes and 01 seconds, and 1 barrage content is released at 1 minute and 25 seconds, it is not difficult to understand that the number of fragment comment contents in the playing time period 0-5 minutes is 4, the number of fragment comment contents in the playing time period 6-10 minutes is 6, and the number of fragment comment contents in the playing time period 11-15 minutes and the playing time period 16-20 minutes are both 0. Correspondingly, if the time information is the display time, then when multimedia data A corresponds to 10 barrage contents whose semantic information is the target semantic information, of which the display time of 3 barrage contents is from 4 minutes 50 seconds to 5 minutes 20 seconds, the release time of 6 barrage contents is from 6 minutes 01 seconds to 6 minutes 31 seconds when the multimedia data is played, and the release time of 1 barrage content is from 9 minutes 25 seconds to 10 minutes 01 seconds, it is not difficult to understand that the number of segment comment contents in the playback time period of 0-5 minutes is 3, the number of segment comment contents in the playback time period of 6-10 minutes is 10 (3+6+1=10), the number of segment comment contents in the playback time period of 11-15 minutes is 1, and the number of segment comment contents in the playback time period of 16-20 minutes is 0.
[0061] S404: Determine the fragment comment content whose quantity meets the second preset quantity condition.
[0062] The second preset quantity condition may include any one of the following: the largest number, the number exceeding the preset quantity threshold, and the number being in the first Q positions after being arranged from large to small (Q is a positive integer, Q is greater than zero and less than or equal to the number of playback time periods). After the topic generation device determines the number of segment comment content that meets the second preset quantity condition, the playback time period of the segment comment content can be used as the time period in which the barrage content appears most densely.
[0063] S405: Determine the starting position of the target multimedia segment matching the discussion topic in the multimedia data according to the time information of the segment comment content with the earliest publishing time among the determined segment comment contents.
[0064] The starting position can be understood as: the specific start time of the target multimedia segment in the playback time of the multimedia data, such as: the target multimedia segment starts playing at the 2nd minute and 30 seconds of the playback time, then the starting position of the target multimedia segment is considered to be: 2 minutes and 30 seconds. In practical applications, since the actual appearance time of the barrage content is not necessarily the start time of the playback time period, in order to obtain a more accurate starting position of the target multimedia segment, the topic generation device can obtain the appearance time of the first segment comment content in the playback time period, wherein the first segment comment content refers to: among all the segment comment content corresponding to the target multimedia segment, the segment comment content with the earliest release time.
[0065] S406: Establish an association relationship between the discussion topic and the starting position, and store the discussion topic and the starting position associated with the discussion topic.
[0066] Among them, the discussion topic and the starting position associated with the discussion topic can be stored on the blockchain, and the blockchain is composed of the terminal device of the user who posts the comment content and one or more servers corresponding to the topic generation device, and the user's terminal device and each of the one or more servers are nodes on the blockchain. It can be understood that by storing the discussion topic in association with the starting position of the target multimedia segment, the topic generation device can trigger the playback of the target multimedia segment or trigger the display of the playback entrance of the target multimedia segment when detecting that the user browses the content related to the discussion topic.
[0067] The topic generation method provided in the embodiment of the present application determines the target multimedia segment that matches the discussion topic in the multimedia data, and stores the discussion topic in association with the starting position of the target multimedia segment, so that the user can quickly jump to the multimedia segment at the corresponding starting position on the discussion page where the discussion topic is located for viewing, which can improve the user experience to a certain extent.
[0068] See also Figure 5 , Figure 5 It is a schematic flow chart of another method for determining a target multimedia segment based on a discussion topic provided by an embodiment of the present application, which can be applied to a scenario where one or more target comment contents are comment contents in a comment area, or one or more target comment contents are bullet screen contents; Figure 5 As shown, in this case, when the topic generation device associates the discussion topic with the target multimedia segment, the steps specifically include steps S501 to S503:
[0069] S501: Determine a target multimedia segment matching a discussion topic in multimedia data.
[0070] In one embodiment, it is not difficult to understand that the duration of the multimedia segment in which the target comment content appears most densely can be longer than the duration corresponding to each playback time period. Therefore, in order to obtain a complete multimedia segment that is strongly related to the discussion topic, when the target comment content is the bullet content, the topic generation device determines the target multimedia segment in a manner that may include the following steps: (1) obtaining the playback duration of the multimedia data; (2) dividing the playback duration into time periods to obtain multiple playback time periods; (3) obtaining the number of segment comment contents of the multimedia segment played in each of the multiple playback time periods; wherein the multimedia segment played in each playback time period is the multimedia segment of the multimedia data, and the semantic information of the segment comment content is the target semantic information; (4) determining the first multimedia segment corresponding to the segment comment content whose number meets the second preset number condition; the first multimedia segment may be: a multimedia segment whose number of bullet content with the same target semantic information meets the second preset number threshold. (5) determining a second multimedia segment adjacent to the first multimedia segment in the multimedia data, wherein the second multimedia segment matches the discussion topic corresponding to the first multimedia segment. (6) obtaining the target multimedia segment based on the first multimedia segment and the second multimedia segment.
[0071] It should be noted that the relevant embodiments of steps (1) to (4) can be found in Figure 4 The relevant description of steps S401 to S404 in the above application will not be repeated here. In addition, the above-mentioned second multimedia segment matches the discussion topic and can be understood as: the second multimedia segment includes the features of each keyword corresponding to the discussion topic (such as: facial features, action features, etc.); then, further, the second multimedia segment can be understood as: in the multimedia data, except for the first multimedia segment, other multimedia segments that include the features of each keyword corresponding to the discussion topic. The playback time of the second multimedia segment may be earlier than the playback time of the first multimedia segment, that is, the second multimedia segment is: a multimedia segment corresponding to a playback time period before the playback time period of the first multimedia segment; or the playback time of the second multimedia segment may be later than the playback time of the first multimedia segment, that is, the second multimedia segment is: a multimedia segment corresponding to a playback time period after the playback time period of the first multimedia segment; or the playback time of the first multimedia sub-segment in the second multimedia segment may be earlier than the playback time of the first multimedia segment, and the playback time of the second multimedia sub-segment in the second multimedia segment is later than the playback time of the first multimedia segment, that is, the second multimedia segment includes both multimedia segments whose playback time period is before the playback time period of the first multimedia segment and multimedia segments whose playback time period is after the playback time period of the first multimedia segment. This application does not impose any limitation on this.
[0072] In another embodiment, when the target comment content is the comment content of the comment area, the topic generation device can determine the target multimedia segment in the following manner: the topic generation device first performs word segmentation processing on the discussion topic to obtain one or more keywords; and then determines the target multimedia segment matching the one or more keywords in the multimedia data. In practical applications, when the topic generation device determines the target multimedia segment matching the one or more keywords in the multimedia data, it can specifically include the following steps: obtaining the recognition material of each keyword in the one or more keywords, and obtaining the material features of each recognition material, wherein the recognition material is used to indicate the features corresponding to the keywords, and the recognition material is used to indicate the features corresponding to the keywords; the topic generation device can recognize each recognition material and obtain the material features of the recognition material; further, after the topic generation device obtains the material features of each recognition material, it can perform feature recognition on the multimedia data based on the material features of each recognition material, and determine a third multimedia segment in the multimedia data, the third multimedia segment includes the material features of the recognition material of each keyword in the one or more keywords, and after determining the third multimedia segment, the topic generation device can use the third multimedia segment as the target multimedia segment matching the one or more keywords.
[0073] Among them, the recognition material may include but is not limited to: facial material (such as: XX's facial image), action material (such as: a video including breakdancing movements, a gif including waving movements, etc.), audio material (such as: the accompaniment of XX song, the sound of XX animal, etc.), text material (the lyrics of XX song, the name of XX character, etc.), static physical material (such as: the image of XX flower, the image of mountain, the image of field, the image of train station, etc.), etc. The topic generation device can use different recognition methods to perform feature recognition for different recognition materials. The following is an explanation of the recognition methods of several common recognition materials: (1) If the recognition material is facial material (such as: the facial image of the main character mentioned in the multimedia data), then the topic generation device can first perform facial recognition on the video data (a type of multimedia data) based on the facial material to obtain the facial features of the facial material, and then determine the video clip including the character indicated by the facial features based on the facial features. Exemplarily, the specific process of the topic generation device performing facial recognition can be found in Figure 6aAs shown, the topic generation device first pre-processes the face material and each face image summarized based on the video data, and then extracts features from the face material and each face image in the video data, and then compares the features of the face material with the features of the face image to determine the face image that matches the features of the face material. For example, the topic generation device can use a feature extraction method based on deep learning (such as: using a convolution neural network (CNN) algorithm) to extract features from the face material. Among them, the so-called CNN is an artificial neural network based on deep learning theory. It mainly uses weight sharing to reduce the parameter expansion problem in ordinary neural networks and uses convolution kernels to perform convolution operations on input data during forward calculation. The result is passed through a nonlinear function as the output of the layer. Such a layer is called a convolution layer. A downsampling layer will appear between the convolution layer and the convolution layer. The downsampling layer is mainly used to obtain the invariance of local features and reduce the scale of the feature space. Generally, after the convolution layer and the downsampling layer, there is a fully connected neural network for final recognition. (2) If the identified material is action material (such as a breakdancing action video, a waving action video, etc.), the topic generation device can first perform action recognition on the video data based on the action material to obtain the action features of the action material, and then determine the video segment in the video data that includes the action indicated by the action features based on the action features. For example, the process of feature extraction of action material by the topic generation device can be seen in Figure 6b As shown, the topic generation device first extracts features from the action material, and then extracts features from the video data to obtain feature representations of the action material, and then can perform action recognition and understanding on the video data based on the feature representations. Optionally, the topic generation device can also associate and store the action features corresponding to the action material in the video data and the determined video clips (such as saving them to a database of a server), so that the topic generation device can directly obtain the video clip corresponding to the action from the database when it needs to recognize the action corresponding to the action material next time, thereby improving the processing efficiency of the topic generation device. (3) If the recognition material is a static physical material, the topic generation device can recognize the static physical material by calling an interface for recognizing the physical object. It can be understood that the topic generation device can complete the recognition of the background in the multimedia data based on this method, and then can determine the multimedia clip including a certain background. (4) If the recognition material is a text material (such as the name, alias, nickname of the fans, etc. of the main actor), the topic generation device can perform text recognition on the text material by calling a natural language recognition model, and then determine the multimedia clip including a certain text feature.
[0074] In another embodiment, when the number of one or more target comment contents is one, and the target comment content is a multimedia segment in the multimedia data, the topic generating device can directly compare the multimedia segment with the multimedia data, determine the position of the multimedia segment in the multimedia data, and then use the determined position as the starting position of the target multimedia segment that matches the discussion topic.
[0075] Based on the above description, the specific process of the topic generation device determining the target multimedia segment matching the discussion topic can be exemplified by referring to Figure 7a As shown, the topic generation device can be exemplarily: a server cluster composed of a multimedia data understanding server and a semantic understanding server. Figure 7a As shown, the operator can first upload the multimedia data that needs to generate the discussion topic, as well as the identification materials that may be needed in the process of generating the topic, to the multimedia data understanding server; it is not difficult to understand that since an identification material may have multiple features, such as when the identification material uploaded by the operator is a video clip, the video clip may include faces and character actions; in this case, simply uploading the video clip as the identification material, the multimedia data understanding server does not know whether the video clip is used to extract the content corresponding to the facial features or the content corresponding to the action features in video A, so the operator needs to arbitrarily determine and upload the features that need to be identified by the identification material, and determine whether the identification material is specifically a facial material or an action material. Next, the multimedia data understanding server can perform feature detection based on the identification material and the multimedia data, and associate the time point of the feature corresponding to the identification material with the feature (such as: storing it in the memory of the multimedia data understanding server). Further, after the multimedia data understanding server receives the features of the keywords of the discussion topic sent by the semantic understanding server, it can directly search for the corresponding multimedia clip from the memory of the multimedia data understanding server based on the features of the keywords, and then obtain the target multimedia clip that meets the features of all keywords.
[0076] based on Figure 7a It can be understood that when the target comment content is a bullet screen content, the specific process of the topic generation device determining the target multimedia segment can be exemplarily referred to in Figure 7b When the target comment content is the comment content of the comment area, the specific process of the topic generation device determining the target multimedia segment can be found in Figure 7c or Figure 7d shown. Figure 7b The hot topic server, bullet screen server, semantic understanding server and video understanding server can be integrated into a topic generation device, which includes the functions of each server in the hot topic server, bullet screen server, semantic understanding server and video understanding server; Figure 7c The hot topic server, comment server, semantic understanding server and video understanding server can be integrated into a topic generation device, which includes the functions of each server in the hot topic server, comment server, semantic understanding server and video understanding server. Figure 7b as well as Figure 7c The specific implementation principles of each step in the flowchart shown are Figure 7a The description is consistent with that in Figure 7d The specific implementation principles of each step in the flowchart shown are Figure 3 or Figure 7a Some of the related steps are consistent with those of , so this application will not be repeated here.
[0077] S502: Determine the starting position of the target multimedia segment in the multimedia data.
[0078] The topic generation device may use the time when the features of all keywords of the discussion topic appear simultaneously for the first time as the starting position of the target multimedia segment in the multimedia data.
[0079] S503: Establish an association relationship between the discussion topic and the starting position, and store the discussion topic and the starting position associated with the discussion topic in a memory.
[0080] In one embodiment, the relevant description of step S503 can refer to step S406, and this application will not repeat it here.
[0081] It is understandable that the generated discussion topic can be displayed in the playback interface where the multimedia data is located, or can be displayed in the discussion interface, such as Figure 8As shown, the discussion interface 81 may include one or more discussion topics, and the user may view multiple discussion topics in the discussion interface 81. For example, the user may click on the display area 811 of the discussion topic on the terminal device to trigger the display of the topic details page 82, and the topic details page 82 includes multimedia data 821 corresponding to the discussion topic 812 (assuming that the multimedia data 821 is video data), and one or more comment contents 822; it should be noted that the user may play the target multimedia segment matching the discussion topic 812 by clicking on the play entry 8211 on the terminal device. In a specific implementation, when the terminal device detects a multimedia segment play instruction for the discussion topic 812 (such as: the terminal device detects that the user clicks on the play entry 8211), a multimedia segment play request may be generated, and the multimedia segment play request may be sent to the terminal device. The request to play the multimedia segment may be sent to the topic generation device, wherein the multimedia segment play request may carry the topic identifier of the discussion topic; for example: the terminal device may generate and send a multimedia segment play request carrying the topic identifier of the discussion topic 812 (such as: the name of the discussion topic "XXXX4") to the topic generation device, so that the topic generation device may respond to the multimedia segment play request, search the memory for the starting position associated with the discussion topic corresponding to the topic identifier, and then obtain the target multimedia segment matching the discussion topic, wherein the starting position of the target multimedia segment may be the starting position obtained by the topic generation device in the memory; further, the topic generation device may send the obtained target multimedia segment to the terminal device, so that the terminal device may directly start playing from the starting position of the target multimedia segment in the multimedia data 821. Optionally, the topic generation device may also obtain the ending position of the target multimedia segment, based on which, the terminal device may stop playing the multimedia data after playing to the ending position of the target multimedia segment, or may continue playing the multimedia data after playing to the ending position of the target multimedia segment, and the present application does not impose any restrictions on this.
[0082] The embodiment of the present application associates and stores the target multimedia clip with the discussion topic, so that the user can directly jump to the target multimedia clip that is strongly related to the discussion topic when participating in the discussion topic, which improves the user experience when participating in the topic discussion to a certain extent.
[0083] Based on the description of the above topic generation method embodiment, the present application embodiment also discloses a topic generation device, which can be a computer program (including program code) running on the above-mentioned server. Figure 1 , Figure 4 or Figure 5 See the method shown in Fig. 9The topic generating device 90 may at least include: a semantic recognition unit 901 and a generating unit 902.
[0084] The semantic recognition unit 901 is used to perform semantic recognition on one or more target comment contents of the multimedia data to obtain target semantic information; wherein the topic degree corresponding to the target comment content or the target semantic information satisfies a preset topic condition, and the topic degree is used to indicate the degree of participation in the multimedia data;
[0085] The generating unit 902 is configured to generate a discussion topic for the multimedia data based on the target semantic information.
[0086] In one implementation, the semantic recognition unit 901 may be specifically used for:
[0087] Performing semantic recognition on the one or more target comment contents to obtain one or more pieces of semantic information;
[0088] Obtaining the number of comment contents corresponding to similar semantic information in the one or more pieces of semantic information;
[0089] If the number of comment contents corresponding to the target similar semantic information satisfies the first preset number condition, determining that the topic degree corresponding to the target similar semantic information satisfies the preset topic condition;
[0090] The target similar semantic information is used as the target semantic information.
[0091] In yet another embodiment, the semantic recognition unit 901 may be specifically used for:
[0092] Obtaining interactive information of each comment content in one or more comment contents of the multimedia data;
[0093] Determining the topicality of each comment content according to the interactive information of each comment content;
[0094] Determining, from the one or more comment contents, target comment contents whose topicality satisfies the preset topic condition;
[0095] Performing semantic recognition on the target comment content to obtain semantic information of the target comment content;
[0096] The semantic information of the target comment content is used as the target semantic information.
[0097] In yet another embodiment, the topic generating device further includes a processing unit 903, wherein the processing unit 903 is configured to:
[0098] Obtaining the playback duration of the multimedia data;
[0099] Dividing the playing time into time periods to obtain multiple playing time periods;
[0100] Acquire the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods, wherein the multimedia segments played in each of the playback time periods are multimedia segments of the multimedia data, and the semantic information of the segment comment contents is the target semantic information;
[0101] Determine the number of segment comment contents that meet the second preset number condition, and determine the starting position of the target multimedia segment matching the discussion topic in the multimedia data according to the time information of the segment comment content with the earliest publishing time among the determined segment comment contents;
[0102] An association relationship between the discussion topic and the starting position is established, and the discussion topic and the starting position associated with the discussion topic are stored in a memory.
[0103] In yet another embodiment, the processing unit 903 is further configured to:
[0104] Determining a target multimedia segment matching the discussion topic in the multimedia data;
[0105] Determining a starting position of the target multimedia segment in the multimedia data;
[0106] An association relationship between the discussion topic and the starting position is established, and the discussion topic and the starting position associated with the discussion topic are stored in a memory.
[0107] In yet another implementation, the processing unit 903 may also be configured to:
[0108] receiving a multimedia segment playing request sent by a terminal device, the multimedia segment playing request carrying a topic identifier of the discussion topic, the multimedia segment playing request being sent by the terminal device when detecting a multimedia segment playing instruction for the discussion topic;
[0109] Searching the memory for a starting position associated with the discussion topic corresponding to the topic identifier;
[0110] Acquire a target multimedia segment that matches the discussion topic, wherein the starting position of the target multimedia segment in the multimedia data corresponding to the discussion topic is the found starting position;
[0111] The target multimedia segment is sent to the terminal device, so that the terminal device plays the target multimedia segment.
[0112] In yet another implementation, the processing unit 903 may also be configured to execute:
[0113] Obtaining the playback duration of the multimedia data;
[0114] Dividing the playing time into time periods to obtain multiple playing time periods;
[0115] Acquire the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods, wherein the multimedia segments played in each of the playback time periods are multimedia segments of the multimedia data, and the semantic information of the segment comment contents is the target semantic information;
[0116] Determine the first multimedia segments corresponding to the segment comment contents whose quantity meets the second preset quantity condition;
[0117] Determining a second multimedia segment adjacent to the first multimedia segment in the multimedia data, wherein the second multimedia segment matches the discussion topic corresponding to the first multimedia segment; wherein the playing time of the second multimedia segment is earlier than the playing time of the first multimedia segment, or the playing time of the second multimedia segment is later than the playing time of the first multimedia segment, or the playing time of the first multimedia sub-segment in the second multimedia segment is earlier than the playing time of the first multimedia segment, and the playing time of the second multimedia sub-segment in the second multimedia segment is later than the playing time of the first multimedia segment;
[0118] The target multimedia segment is obtained based on the first multimedia segment and the second multimedia segment.
[0119] In yet another embodiment, when acquiring the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods, the processing unit 903 may be configured to:
[0120] Acquire time information of each target comment content in the one or more target comment contents during the playing process of the multimedia data;
[0121] Determine, according to the time information of each target comment content and each playback time period, the segment comment content of the multimedia segment played in each playback time period, wherein the segment comment content is part or all of the comment content in the one or more target comment contents;
[0122] The number of segment comment contents of the multimedia segments played in each playing time period is obtained.
[0123] According to one embodiment of the present application, Figure 1 and Figure 4 and Figure 5 The various steps involved in the method shown can be performed by Fig. 9 The topic generation device 90 shown in FIG. Figure 1 Step S101 shown can be performed by Fig. 9 The topic generation device 90 shown in FIG. 1 is implemented by the semantic recognition unit 901, and step S102 can be performed by Fig. 9 The topic generating unit 902 in the topic generating device 90 shown in FIG. Figure 4 Steps S401 to S406 shown in FIG. 4 can be performed by Fig. 9 The processing unit 903 in the topic generating device 90 shown in the figure is executed; for example, Figure 5 Steps S501 to S503 shown in FIG. Fig. 9 The topic generation device 90 shown in the figure is executed by the processing unit 903.
[0124] According to another embodiment of the present application, Fig. 9 The various units in the topic generation device 90 shown are divided based on logical functions, and the above-mentioned various units can be separately or all combined into one or several other units to constitute, or one (some) of the units can be further divided into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. In other embodiments of the present application, the above-mentioned topic generation device can also include other units. In actual applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.
[0125] According to another embodiment of the present application, the program can be executed by running on a general computing device such as a computer including a central processing unit (CPU), a random access memory medium (RAM), a read-only memory medium (ROM), and other processing elements and storage elements. Figure 1 or Figure 4 or Figure 5 A computer program (including program code) for each step of the method shown in FIG. Fig. 9 The topic generating device 90 shown in the figure can realize the topic generating method of the embodiment of the present application. The computer program can be recorded on a computer storage medium, for example, and loaded into the above-mentioned computer device through the computer storage medium and run therein.
[0126] The topic generation device provided in the embodiment of the present application performs semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information, and then generates discussion topics for multimedia data based on the target semantic information. Since the discussion topics are generated after semantic understanding of one or more target comment contents, the subjectivity of manually editing the discussion topics is avoided, and the efficiency of topic generation can be improved to a certain extent. In addition, since exclusive discussion topics are generated for discussion hotspots, the multimedia data can have more topic volume.
[0127] Based on the description of the above method embodiment and the device embodiment, the present application embodiment also provides a topic generation device, see Fig.10 The topic generating device 100 at least includes a processor 1001 and a computer storage medium 1002, and the processor 1001 and the computer storage medium 1002 can be connected via a bus or other means.
[0128] The computer storage medium 1002 is a memory device in a computer device for storing programs and data. It is understandable that the computer storage medium 1002 here can include both the built-in storage medium in the topic generation device and the extended storage medium supported by the topic generation device. The computer storage medium 1002 provides a storage space, which stores the operating system of the topic generation device. In addition, one or more instructions suitable for being loaded and executed by the processor 1001 are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory; optionally, it can also be at least one computer storage medium located away from the aforementioned processor. The processor 1001 (or CPU (Central Processing Unit)) is the computing core and control core of the topic generation device, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to realize the corresponding method flow or corresponding function.
[0129] In one embodiment, the processor 1001 may load and execute one or more instructions stored in the computer storage medium 1002 to implement the above-mentioned Figure 1 and Figure 4 and Figure 5 Corresponding method steps in the method embodiment shown; in a specific implementation, one or more instructions in the computer storage medium 1002 are loaded by the processor 1001 and execute the following steps:
[0130] Performing semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information; wherein the topic degree corresponding to the target comment content or the target semantic information satisfies a preset topic condition, and the topic degree is used to indicate the degree of participation in the multimedia data;
[0131] Based on the target semantic information, a discussion topic of the multimedia data is generated.
[0132] In one implementation, the processor 1001 may be specifically configured to:
[0133] Performing semantic recognition on the one or more target comment contents to obtain one or more pieces of semantic information;
[0134] Obtaining the number of comment contents corresponding to similar semantic information in the one or more pieces of semantic information;
[0135] If the number of comment contents corresponding to the target similar semantic information satisfies the first preset number condition, determining that the topic degree corresponding to the target similar semantic information satisfies the preset topic condition;
[0136] The target similar semantic information is used as the target semantic information.
[0137] In yet another implementation, the processor 1001 may also be configured to execute:
[0138] Obtaining interactive information of each comment content in one or more comment contents of the multimedia data;
[0139] Determining the topicality of each comment content according to the interactive information of each comment content;
[0140] Determining, from the one or more comment contents, target comment contents whose topicality satisfies the preset topic condition;
[0141] Performing semantic recognition on the target comment content to obtain semantic information of the target comment content;
[0142] The semantic information of the target comment content is used as the target semantic information.
[0143] In yet another implementation, the processor 1001 may be specifically configured to:
[0144] Obtaining the playback duration of the multimedia data;
[0145] Dividing the playing time into time periods to obtain multiple playing time periods;
[0146] Acquire the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods, wherein the multimedia segments played in each of the playback time periods are multimedia segments of the multimedia data, and the semantic information of the segment comment contents is the target semantic information;
[0147] Determine the number of segment comment contents that meet the second preset number condition, and determine the starting position of the target multimedia segment matching the discussion topic in the multimedia data according to the time information of the segment comment content with the earliest publishing time among the determined segment comment contents;
[0148] An association relationship between the discussion topic and the starting position is established, and the discussion topic and the starting position associated with the discussion topic are stored in a memory.
[0149] In yet another implementation, the processor 1001 may be specifically configured to:
[0150] Determining a target multimedia segment matching the discussion topic in the multimedia data;
[0151] Determining a starting position of the target multimedia segment in the multimedia data;
[0152] An association relationship between the discussion topic and the starting position is established, and the discussion topic and the starting position associated with the discussion topic are stored in a memory.
[0153] In yet another implementation, the processor 1001 may be specifically configured to:
[0154] receiving a multimedia segment playing request sent by a terminal device, the multimedia segment playing request carrying a topic identifier of the discussion topic, the multimedia segment playing request being sent by the terminal device when detecting a multimedia segment playing instruction for the discussion topic;
[0155] Searching the memory for a starting position associated with the discussion topic corresponding to the topic identifier;
[0156] Acquire a target multimedia segment that matches the discussion topic, wherein the starting position of the target multimedia segment in the multimedia data corresponding to the discussion topic is the found starting position;
[0157] The target multimedia segment is sent to the terminal device, so that the terminal device plays the target multimedia segment.
[0158] In yet another implementation, the processor 1001 may be specifically configured to:
[0159] Obtaining the playback duration of the multimedia data;
[0160] Dividing the playing time into time periods to obtain multiple playing time periods;
[0161] Acquire the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods, wherein the multimedia segments played in each of the playback time periods are multimedia segments of the multimedia data, and the semantic information of the segment comment contents is the target semantic information;
[0162] Determine the first multimedia segments corresponding to the segment comment contents whose quantity meets the second preset quantity condition;
[0163] Determining a second multimedia segment adjacent to the first multimedia segment in the multimedia data, wherein the second multimedia segment matches the discussion topic corresponding to the first multimedia segment; wherein the playing time of the second multimedia segment is earlier than the playing time of the first multimedia segment, or the playing time of the second multimedia segment is later than the playing time of the first multimedia segment, or the playing time of the first multimedia sub-segment in the second multimedia segment is earlier than the playing time of the first multimedia segment, and the playing time of the second multimedia sub-segment in the second multimedia segment is later than the playing time of the first multimedia segment;
[0164] The target multimedia segment is obtained based on the first multimedia segment and the second multimedia segment.
[0165] In yet another implementation, the processor 1001 may be specifically configured to:
[0166] Acquire time information of each target comment content in the one or more target comment contents during the playing process of the multimedia data;
[0167] Determine, according to the time information of each target comment content and each playback time period, the segment comment content of the multimedia segment played in each playback time period, wherein the segment comment content is part or all of the comment content in the one or more target comment contents;
[0168] The number of segment comment contents of the multimedia segments played in each playing time period is obtained.
[0169] The topic generation device provided in the embodiment of the present application performs semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information, and then generates discussion topics of multimedia data based on the target semantic information. Since the discussion topics are generated after semantic understanding of one or more target comment contents, the subjectivity of manual editing of discussion topics is avoided, and the efficiency of topic generation can be improved to a certain extent. In addition, since exclusive discussion topics are generated for discussion hotspots, the multimedia data can have more topic volume.
[0170] The embodiment of the present application also provides a computer storage medium, in which a computer program of the above-mentioned topic generation method is stored, and the computer program includes program instructions. When one or more processors load and execute the program instructions, the description of the topic generation method in the embodiment can be implemented, which will not be repeated here. The description of the beneficial effects of adopting the same method will not be repeated here. It is understandable that the program instructions can be deployed on one or more devices that can communicate with each other for execution.
[0171] It should be noted that, according to one aspect of the present application, a computer program product or a computer program is also provided, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor in the topic generation device reads the computer instructions from the computer-readable storage medium and then executes the computer instructions, thereby enabling the topic generation device to perform the above Figure 1 and Figure 4 and Figure 5 The method for topic generation is provided in various optional aspects of the embodiment of the method.
[0172] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment method can be implemented by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiment of the image processing method. The computer-readable storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).
[0173] What is disclosed above is only a partial embodiment of the present application, which certainly cannot be used to limit the scope of rights of the present application. A person skilled in the art can understand that the entire or partial process of implementing the above embodiment and the equivalent changes made according to the claims of the present application are still within the scope of the invention.
Claims
1. A topic generation method, characterized in that: include: Performing semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information; wherein the topic degree corresponding to the target comment content or the target semantic information satisfies a preset topic condition, and the topic degree is used to indicate the degree of participation in the multimedia data; Based on the target semantic information, generating a discussion topic for the multimedia data; Dividing the obtained playback time of the multimedia data into time periods to obtain multiple playback time periods; Acquire the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods, wherein the multimedia segments played in each of the playback time periods are multimedia segments of the multimedia data, and the semantic information of the segment comment contents is the target semantic information; Determine the number of segment comment contents that meet the second preset number condition, and determine the starting position of the target multimedia segment matching the discussion topic in the multimedia data according to the time information of the segment comment content with the earliest publishing time among the determined segment comment contents; An association relationship between the discussion topic and the starting position is established, and the discussion topic and the starting position associated with the discussion topic are stored in a memory.
2. The method according to claim 1, characterized in that The performing semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information includes: Performing semantic recognition on the one or more target comment contents to obtain one or more pieces of semantic information; Obtaining the number of comment contents corresponding to similar semantic information in the one or more pieces of semantic information; If the number of comment contents corresponding to the target similar semantic information satisfies the first preset number condition, determining that the topic degree corresponding to the target similar semantic information satisfies the preset topic condition; The target similar semantic information is used as the target semantic information.
3. The method according to claim 1, characterized in that The performing semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information includes: Obtaining interactive information of each comment content in one or more comment contents of the multimedia data; Determining the topicality of each comment content according to the interactive information of each comment content; Determining, from the one or more comment contents, target comment contents whose topicality satisfies the preset topic condition; Performing semantic recognition on the target comment content to obtain semantic information of the target comment content; The semantic information of the target comment content is used as the target semantic information.
4. The method according to claim 1, characterized in that: The method further comprises: Determining a target multimedia segment matching the discussion topic in the multimedia data; Determining a starting position of the target multimedia segment in the multimedia data; An association relationship between the discussion topic and the starting position is established, and the discussion topic and the starting position associated with the discussion topic are stored in a memory.
5. The method according to claim 1 or 4, characterized in that: The method further comprises: receiving a multimedia segment playing request sent by a terminal device, the multimedia segment playing request carrying a topic identifier of the discussion topic, the multimedia segment playing request being sent by the terminal device when detecting a multimedia segment playing instruction for the discussion topic; Searching the memory for a starting position associated with the discussion topic corresponding to the topic identifier; Acquire a target multimedia segment that matches the discussion topic, wherein the starting position of the target multimedia segment in the multimedia data corresponding to the discussion topic is the found starting position; The target multimedia segment is sent to the terminal device, so that the terminal device plays the target multimedia segment.
6. The method according to claim 4, characterized in that The step of determining a target multimedia segment matching the discussion topic in the multimedia data includes: Obtaining the playback duration of the multimedia data; Dividing the playing time into time periods to obtain multiple playing time periods; Acquire the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods, wherein the multimedia segments played in each of the playback time periods are multimedia segments of the multimedia data, and the semantic information of the segment comment contents is the target semantic information; Determine the first multimedia segments corresponding to the segment comment contents whose quantity meets the second preset quantity condition; Determining a second multimedia segment adjacent to the first multimedia segment in the multimedia data, wherein the second multimedia segment matches the discussion topic corresponding to the first multimedia segment; wherein the playing time of the second multimedia segment is earlier than the playing time of the first multimedia segment, or the playing time of the second multimedia segment is later than the playing time of the first multimedia segment, or the playing time of the first multimedia sub-segment in the second multimedia segment is earlier than the playing time of the first multimedia segment, and the playing time of the second multimedia sub-segment in the second multimedia segment is later than the playing time of the first multimedia segment; The target multimedia segment is obtained based on the first multimedia segment and the second multimedia segment.
7. The method according to claim 1 or 6, characterized in that: The obtaining of the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods includes: Acquire time information of each target comment content in the one or more target comment contents during the playing process of the multimedia data; Determine, according to the time information of each target comment content and each playback time period, the segment comment content of the multimedia segment played in each playback time period, wherein the segment comment content is part or all of the comment content in the one or more target comment contents; The number of segment comment contents of the multimedia segments played in each playing time period is obtained.
8. A topic generating device, characterized in that: include: A semantic recognition unit, configured to perform semantic recognition on one or more target comment contents of multimedia data to obtain target semantic information; wherein the topic degree corresponding to the target comment content or the target semantic information satisfies a preset topic condition, and the topic degree is used to indicate the degree of participation in the multimedia data; A generating unit, configured to generate a discussion topic for the multimedia data based on the target semantic information; A processing unit is used to obtain the playback time of the multimedia data; divide the playback time into time periods to obtain multiple playback time periods; obtain the number of segment comment contents of the multimedia segments played in each of the multiple playback time periods, wherein the multimedia segments played in each playback time period are multimedia segments of the multimedia data, and the semantic information of the segment comment contents is the target semantic information; determine the segment comment contents whose number meets the second preset number condition, and determine the starting position of the target multimedia segment matching the discussion topic in the multimedia data according to the time information of the segment comment contents with the earliest publishing time in the determined segment comment contents; establish an association relationship between the discussion topic and the starting position, and store the discussion topic and the starting position associated with the discussion topic in a memory.
9. A computer storage medium, characterized in that The computer storage medium stores one or more instructions, and the one or more instructions are suitable for being loaded by a processor and executing the topic generating method according to any one of claims 1 to 7.
10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the topic generation method as described in any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
Video tag extraction method based on semantic association of barrages
CN106407484A
Method, device, computer device and storage medium for generating topic between live broadcast room
CN109271509A
