WO2021190174A1 - 信息确定方法、装置、计算机设备及存储介质 - Google Patents

信息确定方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2021190174A1
WO2021190174A1 PCT/CN2021/075270 CN2021075270W WO2021190174A1 WO 2021190174 A1 WO2021190174 A1 WO 2021190174A1 CN 2021075270 W CN2021075270 W CN 2021075270W WO 2021190174 A1 WO2021190174 A1 WO 2021190174A1
Authority
WO
WIPO (PCT)
Prior art keywords
multimedia resource
information
target
target multimedia
resource
Prior art date
Application number
PCT/CN2021/075270
Other languages
English (en)
French (fr)
Inventor
刘刚
Original Assignee
腾讯科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 腾讯科技(深圳)有限公司 filed Critical 腾讯科技(深圳)有限公司
Publication of WO2021190174A1 publication Critical patent/WO2021190174A1/zh
Priority to US17/721,295 priority Critical patent/US12001474B2/en

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/43Querying
    • G06F16/435Filtering based on additional data, e.g. user or group profiles
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/45Clustering; Classification
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00Computing arrangements using knowledge-based models
    • G06N5/02Knowledge representation; Symbolic representation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods

Definitions

  • the embodiments of the present application relate to the field of computer technology, and in particular, to an information determination method, device, computer equipment, and storage medium.
  • the multimedia resources disseminated on the Internet are becoming more and more abundant.
  • a user when a user is viewing a multimedia resource, he can also publish information for the multimedia resource.
  • other users When other users are viewing the multimedia resource, they can view the information published by the user.
  • the embodiments of the present application provide an information determination method, device, computer equipment, and storage medium, which can improve the accuracy of content information determination.
  • the technical solution is as follows:
  • a method for determining information includes:
  • information matching the target multimedia resource is selected from the multiple pieces of information, and the selected information is determined as the information of the target multimedia resource.
  • an information determining device including:
  • the feature vector acquisition module is used to acquire the feature vector of the target multimedia resource and the feature vector of multiple candidate multimedia resources;
  • a multimedia resource selection module configured to select, from the multiple candidate multimedia resources, a reference multimedia resource whose feature vector matches the feature vector of the target multimedia resource;
  • the information determining module is configured to select information matching the target multimedia resource from the multiple pieces of information according to the multiple pieces of information of the reference multimedia resource, and determine the selected information as the information of the target multimedia resource.
  • the feature vector acquiring module includes:
  • the graph network creation unit is configured to create a graph network based on the target multimedia resource and the multiple candidate multimedia resources, the graph network including the target multimedia resource node corresponding to the target multimedia resource and the multiple candidates A plurality of candidate multimedia resource nodes corresponding to the multimedia resource, and any two multimedia resource nodes that meet the first association condition are connected;
  • the first feature vector obtaining unit is configured to process the graph network based on the first feature extraction model, obtain the feature vector of the target multimedia resource node and the feature vector of the multiple candidate multimedia resource nodes, and convert the The feature vector of the target multimedia resource node is used as the feature vector of the target multimedia resource, and the feature vectors of the multiple candidate multimedia resource nodes are used as the feature vector of the multiple candidate multimedia resources.
  • the graph network creation unit is further configured to perform word segmentation processing on the text resources in the target multimedia resources and the text resources in the multiple candidate multimedia resources to obtain multiple first words;
  • the target multimedia resource, the plurality of candidate multimedia resources, and the plurality of first words create a graph network, the graph network including the target multimedia resource node corresponding to the target multimedia resource, and the plurality of candidate multimedia resources
  • a plurality of candidate multimedia resource nodes corresponding to the resource and a plurality of word nodes corresponding to the plurality of first words, and the word nodes meeting the second association condition are connected to the multimedia resource node.
  • the device further includes:
  • the node determination module is configured to determine that the word node corresponding to the first word and the multimedia resource node corresponding to the multimedia resource satisfy The second association condition.
  • the multimedia resources include video resources;
  • the feature vector acquisition module includes:
  • the frame extraction processing unit is configured to perform frame extraction processing on the target multimedia resource and the multiple candidate multimedia resources respectively to obtain multiple video frames corresponding to the target multimedia resource and the multiple candidate multimedia resources correspondence Multiple video frames;
  • the second feature vector acquiring unit is configured to process multiple video frames corresponding to the target multimedia resource and multiple video frames corresponding to the multiple candidate multimedia resources, respectively, based on the second feature extraction model, to obtain the The feature vector of the target multimedia resource and the feature vector of the multiple candidate multimedia resources.
  • the second feature vector obtaining unit is further configured to separately process multiple video frames corresponding to any multimedia resource based on the second feature extraction model to obtain feature vectors of the multiple video frames Merging the feature vectors of the multiple video frames to obtain the feature vector of the multimedia resource.
  • the multimedia resource selection module includes:
  • the first matching degree obtaining unit is configured to obtain the matching degree between each candidate multimedia resource and the target multimedia resource according to the feature vector of the target multimedia resource and the feature vector of the multiple candidate multimedia resources;
  • the multimedia resource selection unit is configured to select a reference multimedia resource from the multiple candidate multimedia resources according to the degree of matching between each candidate multimedia resource and the target multimedia resource, the reference multimedia resource and the target The degree of matching of the multimedia resources is greater than the degree of matching of other candidate multimedia resources with the target multimedia resources.
  • the information determining module includes:
  • the second matching degree obtaining unit is configured to obtain the matching degree between each of the multiple pieces of information and the target multimedia resource according to multiple pieces of information of the reference multimedia resource;
  • the information selecting unit is configured to select the information of the target multimedia resource from the multiple pieces of information according to the degree of matching between each information and the target multimedia resource, and the information of the target multimedia resource is consistent with the target multimedia
  • the matching degree of the resource is greater than the matching degree of other information with the target multimedia resource.
  • the second matching degree obtaining unit is further configured to perform word segmentation processing on any information to obtain a plurality of second words; and obtain all the second words according to the frequency of each second word in the target multimedia resource.
  • the degree of matching between each of the second words and the target multimedia resource; according to the weights of the plurality of second words, the degree of matching between the plurality of second words and the target multimedia resource is weighted to obtain the The degree of matching between the information and the target multimedia resource.
  • the second matching degree obtaining unit is further configured to determine the degree of matching between each piece of information and the target multimedia resource according to the feature vector of each piece of information and the feature vector of the target multimedia resource .
  • the information determining module includes:
  • the information determining unit is configured to select information belonging to the target classification label and matching the target multimedia resource from the plurality of information according to the plurality of information and the classification label of each information, and determine the selected information as The information of the target multimedia resource.
  • the device further includes:
  • the classification label determination module is configured to process the multiple pieces of information based on the classification model, and determine the classification label of each piece of information.
  • a computer device in another aspect, includes a processor and a memory, and at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor, so as to implement the above-mentioned aspect The described information determination method.
  • a computer-readable storage medium is provided, and at least one instruction is stored in the computer-readable storage medium, and the at least one instruction is loaded and executed by a processor, so as to realize the information determination as described in the above-mentioned aspect. method.
  • the method, device, computer equipment, and storage medium provided by the embodiments of the present application obtain the feature vector of the target multimedia resource and multiple candidate multimedia resources, and select the feature vector and the feature vector of the target multimedia resource from the multiple candidate multimedia resources Matching reference multimedia resources, so that the obtained reference multimedia resources can be matched with the target multimedia resources.
  • the information that matches the target multimedia resources is selected from the multiple information, and the selected information is selected.
  • the information determined as the target multimedia resource provides a way to automatically determine the information for the target multimedia resource, which can ensure that the determined information matches the target multimedia resource and improve the accuracy of the information.
  • FIG. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present application.
  • FIG. 2 is a flowchart of an information determination method provided by an embodiment of the present application.
  • FIG. 3 is a flowchart of an information determination method provided by an embodiment of the present application.
  • FIG. 4 is a flowchart of an information determination method provided by an embodiment of the present application.
  • FIG. 5 is a flowchart of an information determination method provided by an embodiment of the present application.
  • Fig. 6 is a schematic structural diagram of an information management system provided by an embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of an information management system provided by an embodiment of the present application.
  • FIG. 8 is a schematic structural diagram of an information determining device provided by an embodiment of the present application.
  • FIG. 9 is a schematic structural diagram of an information determination device provided by an embodiment of the present application.
  • FIG. 10 is a schematic structural diagram of a terminal provided by an embodiment of the present application.
  • FIG. 11 is a schematic structural diagram of a server provided by an embodiment of the present application.
  • first the first threshold
  • second the second threshold
  • first threshold the first threshold
  • At least one includes one, two or more than two, and multiple includes two or more than two, and each One refers to each of the corresponding multiple, and any one refers to any one of the multiple.
  • multiple elements include 3 elements, and each refers to each of these 3 elements, any one refers to any one of these 3 elements, which can be the first or the second One or the third one.
  • Feeds (a form of information): The website disseminates information to users in the form of Feeds, arranged in a timeline manner. Timeline is the most primitive, intuitive and basic display form of Feeds.
  • AI Artificial Intelligence
  • digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.
  • artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a similar way to human intelligence.
  • Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.
  • Artificial intelligence technology is a comprehensive discipline, covering a wide range of fields, including both hardware-level technology and software-level technology.
  • Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, and mechatronics.
  • Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning/deep learning.
  • Natural language processing (Nature Language Processing, NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that enable effective communication between humans and computers in natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language people use daily, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question answering, knowledge graph and other technologies.
  • Machine Learning is a multi-field interdisciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other subjects. Specializing in the study of how computers simulate or realize human learning behaviors in order to acquire new knowledge or skills, and reorganize the existing knowledge structure to continuously improve its own performance.
  • Machine learning is the core of artificial intelligence, the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence.
  • Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, teaching learning and other technologies.
  • Deep learning The concept of deep learning is derived from the research of artificial neural networks. Multilayer perceptrons containing multiple hidden layers are a kind of deep learning structure. Deep learning combines low-level features to form more abstract high-level representation attribute categories or features to discover distributed feature representations of data.
  • the solution provided by the embodiment of the application based on artificial intelligence machine learning technology, trains the first feature extraction model and the second feature extraction model, and uses the trained first feature extraction model and the second feature extraction model to obtain the features of the multimedia resource Vector, the subsequent implementation determines the content information for the target multimedia resource.
  • FIG. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present application.
  • the implementation environment includes a terminal 101 and a server 102.
  • the terminal 101 establishes a communication connection with the server 102, and interacts through the established communication connection.
  • the terminal 101 is a variety of types of terminals such as mobile phones, computers, and tablet computers.
  • the server 102 is a server, or a server cluster composed of several servers, or a cloud computing server center.
  • the server 102 determines the reference multimedia resource whose feature vector matches the feature vector of the target multimedia resource according to the feature vector of the target multimedia resource and the feature vector of multiple candidate multimedia resources, and selects the target multimedia from multiple information from the reference multimedia resource.
  • the resource matching information is determined to be the information of the target multimedia resource.
  • the method provided in the embodiment of the present application can be used to determine a scene of content information for a multimedia resource.
  • the computer device After the computer device obtains the video resource, it adopts the method for determining the comment information provided in the embodiments of this application to migrate the comment information of other video resources as the comment information of the video resource, so that other users can watch the video resource when watching the video resource.
  • the comment information can be viewed, thereby increasing the number of comment information of the video resource, thereby increasing the popularity of the video resource.
  • the application server After obtaining the article published by the user, the application server adopts the method for determining the comment information provided in the embodiment of this application to migrate multiple comment information of other articles as the comment information of the article, so that the newly published article has multiple comments. Comment information, thereby increasing the amount of comment information for the article.
  • the user selects the article to view by viewing the number of comment information of each article, which improves the attractiveness of the article to the user.
  • the method for determining information provided by the embodiments of the present application is applied to a computer device.
  • the computer device includes a terminal or a server.
  • the terminal is a mobile phone, a computer, a tablet, and other types of terminals.
  • the server is a server or A server cluster composed of several servers, or a cloud computing server center.
  • Fig. 2 is a flowchart of an information determination method provided by an embodiment of the present application, which is applied to a computer device. As shown in Fig. 2, the method includes:
  • a computer device obtains a feature vector of a target multimedia resource and feature vectors of multiple candidate multimedia resources.
  • multimedia resources include video resources, audio resources, text resources, image resources, and so on.
  • the multimedia resource is a resource published by any publisher, for example, a multimedia resource published by an application manager, or a multimedia resource published by a user of the application.
  • the target application is installed in the terminal, the publisher publishes the multimedia resource through the target application, and the terminal uploads the published multimedia resource to the application server corresponding to the target application, and stores it in the application server.
  • the target applications are video applications, music applications, reading applications, and so on.
  • the target multimedia resource and the candidate multimedia resource belong to the same type of multimedia resources.
  • the target multimedia resource and the candidate multimedia resource are both video resources or text resources; or the target multimedia resource and the candidate multimedia resource are different Types of multimedia resources, for example, the target multimedia resource is a text resource, and the candidate multimedia resource is a video resource.
  • any multimedia resource is used as the target multimedia resource, and other multimedia resources are used as candidate multimedia resources, or the multimedia resources that meet the conditions are used as target multimedia resources, and other multimedia resources are used as candidate multimedia resources.
  • the multimedia resource with the amount of information less than the fourth threshold is used as the target multimedia resource
  • the multimedia resource with the amount of information not less than the fourth threshold is used as the candidate multimedia resource.
  • the feature vector is a vector used to represent feature information of the multimedia resource, and the feature vector includes multiple dimensions. Since the feature information of different multimedia resources is different, the feature vectors corresponding to different multimedia resources are different.
  • the following two methods are used to obtain the feature vector of the multimedia resource.
  • the first method includes the following steps 2011-2012:
  • the graph network is a representation of the connection relationship between multiple nodes.
  • the graph network includes a target multimedia resource node corresponding to the target multimedia resource and multiple candidate multimedia resource nodes corresponding to multiple candidate multimedia resources.
  • the multiple multimedia resource nodes connect any two multimedia resource nodes that meet the first association condition to obtain the graph network.
  • the matching degree of any two multimedia resources in response to the matching degree being greater than the fifth threshold, it is determined that the corresponding two multimedia resources correspond to The multimedia resource node meets the first association condition.
  • the matching degree of any two multimedia resources is used to indicate the matching degree between the two multimedia resources.
  • this step 2011 includes the following steps 1-2:
  • Step 1 Perform word segmentation processing on the text resources in the target multimedia resources and the text resources in the multiple candidate multimedia resources to obtain multiple first words.
  • text resources are resources that include text.
  • the text resource of the multimedia resource is the multimedia resource itself; when the multimedia resource is a video resource, the text resource of the multimedia resource includes the title and brief information of the multimedia resource, or includes the video The subtitle information in the resource; when the multimedia resource is an image resource, the text resource of the multimedia resource includes the title and brief information of the multimedia resource, or includes text information in the image resource; when the multimedia resource is an audio resource, the multimedia resource The text resource of includes the title and introduction information of the multimedia resource, or includes the text resource converted from the audio resource.
  • Word segmentation is the process of dividing multiple consecutive words in a text resource into words. Perform word segmentation processing on the text resources in the target multimedia resources, and perform word segmentation processing on the text resources in each candidate multimedia resource, so as to obtain multiple first words after the word segmentation processing of multiple text resources.
  • this step 2011 further includes: performing word segmentation processing on a plurality of candidate multimedia resources and target multimedia resources respectively to obtain a plurality of third words, and performing deduplication processing on the plurality of third words , Get the multiple first words.
  • Step 2 Create a graph network according to the target multimedia resources, multiple candidate multimedia resources, and multiple first words.
  • the graph network includes a target multimedia resource node corresponding to the target multimedia resource, a plurality of candidate multimedia resource nodes corresponding to a plurality of candidate multimedia resources, and a plurality of word nodes corresponding to a plurality of first words.
  • a target multimedia resource node corresponding to the target multimedia resource
  • candidate multimedia resource nodes corresponding to a plurality of candidate multimedia resources
  • word nodes corresponding to a plurality of first words.
  • the text resource in response to the multimedia resource includes the first Words, it is determined that the word node corresponding to the first word and the multimedia resource node corresponding to the multimedia resource satisfy the second association condition.
  • the frequency of occurrence indicates the number of occurrences of the word in the multimedia resource.
  • the first threshold is an arbitrarily set value, such as 5 or 6, etc. The higher the occurrence frequency of a word in the multimedia resource, the higher the matching degree between the word and the multimedia resource, and the lower the occurrence frequency of the word in the multimedia resource, the lower the matching degree between the word and the multimedia resource.
  • the graph network determine the frequency of occurrence of each first word in each multimedia resource. When the frequency of occurrence of any word in any multimedia resource is greater than the first threshold, then the word node corresponding to the word is associated with the multimedia resource. The multimedia resource nodes corresponding to the resources are connected to obtain the graph network.
  • the graph network since among the plurality of first words, different first words have the same meaning, that is, different first words are synonyms. Therefore, when constructing the graph network, by connecting word nodes with the same meaning, the graph network includes the relationship between the word nodes, thereby improving the accuracy of obtaining the graph network.
  • Process the graph network based on the first feature extraction model obtain the feature vector of the target multimedia resource node and the feature vector of multiple candidate multimedia resource nodes, and use the feature vector of the target multimedia resource node as the feature vector of the target multimedia resource.
  • the feature vectors of multiple candidate multimedia resource nodes are used as feature vectors of multiple candidate multimedia resources.
  • the graph network is processed based on the first feature extraction model, and the first feature extraction model outputs the feature vector of each multimedia resource node, thereby obtaining the feature vector of the target multimedia resource node and the feature vector of multiple candidate multimedia resource nodes.
  • the first feature extraction model is a GCN (Graph Convolutional Network) model.
  • the GCN model extracts the features of each node in the graph network, obtains the feature vector of each node through Graph Embedding (graph embedding representation), and processes the graph network based on the GCN model to realize the Node Classification (node classification), Graph Classification (graph classification), Link Prediction (edge prediction).
  • the GCN model is trained through a semi-supervised learning method, and the feature vector of each node in the graph network is obtained based on the trained GCN model; or the GCN model is not trained, and the GCN model based on the initialization parameters is performed on the graph network. Process to get the feature vector of each node.
  • the GCN model is obtained as a DeepWalk (a neural network) model, or a Word2vec (Word To Vector, a word vector model).
  • the first feature extraction model is the GAT (Graph Attention Network, graph attention network) model.
  • the GAT model introduces an Attention mechanism when processing the graph network, and according to the weights between adjacent nodes, The relationship between adjacent nodes is determined, thereby improving the accuracy of fetching the feature vectors of the target multimedia resource node and multiple candidate multimedia resource nodes.
  • the graph network after acquiring the feature vector of each node in the graph network based on the trained GCN model, the graph network can also be used as a sample to continue training the GCN model.
  • the graph network since the graph network includes multiple multimedia resources and the relationship between multiple words, the feature vector of each multimedia resource is subsequently obtained through the graph network, and multiple multimedia resources and multiple words can be incorporated into the feature vector. The relationship between the words, thereby improving the accuracy of the acquired feature vector of the multimedia resource.
  • the second method includes the following steps 2013-2014:
  • the multimedia resources include video resources, and each video resource includes multiple video frames. Therefore, it is necessary to perform frame extraction processing on each video resource to obtain multiple video frames corresponding to each multimedia resource. For example, if the duration of the video resource is 20 seconds and one video frame per second, frame extraction is performed on the video resource to obtain 20 video frames, that is, 20 images are obtained.
  • multiple video frames corresponding to the target multimedia resource and multiple video frames corresponding to multiple candidate multimedia resources are respectively processed to obtain the feature vector of the target multimedia resource and multiple candidate multimedia resources Eigenvectors.
  • the second feature extraction model is a model for acquiring feature vectors of multimedia resources, and is a CNN (Convolutional Neural Networks, convolutional neural network) model or other models.
  • CNN Convolutional Neural Networks, convolutional neural network
  • this step 2014 includes: based on the second feature extraction model, multiple video frames corresponding to any multimedia resource are processed separately to obtain feature vectors of multiple video frames. The feature vector is merged to obtain the feature vector of the multimedia resource.
  • the second feature extraction model includes a feature detection sub-model and a global feature aggregation sub-model; when acquiring the feature vector of any video frame, it is based on the feature
  • the detection sub-model detects the local feature points of the video frame, and obtains the detection results of multiple local features in the video frame.
  • the multiple local features are aggregated to obtain the video frame Feature vector.
  • the degree of matching is used to indicate the degree of matching between the candidate multimedia resource and the target multimedia resource.
  • the greater the degree of matching the more relevant the candidate multimedia resource is to the target multimedia resource, and the smaller the degree of matching indicates the candidate multimedia resource and the target.
  • the more irrelevant multimedia resources are.
  • the degree of matching between the candidate multimedia resource and the target multimedia resource uses the cosine distance, Euclidean distance, Mahalanobis distance, Chebyshev distance, Hamming distance, Standardized Euclidean distance and so on.
  • the similarity (A, B) between the candidate multimedia resource and the target multimedia resource is determined, and the feature The vector A, the feature vector B, and the matching degree similarity (A, B) satisfy the following relationship:
  • is used to represent the length of feature vector A
  • is used to represent the length of feature vector B
  • a i represents the value of the i-th dimension in feature vector A
  • B i represents feature vector B
  • the value of the i-th dimension in, i is a positive integer not less than 1 and not greater than n
  • n is used to indicate the number of dimensions included in the feature vector, and n is a positive integer greater than or equal to 2.
  • the computer device selects a reference multimedia resource from a plurality of candidate multimedia resources according to the degree of matching between each candidate multimedia resource and the target multimedia resource.
  • the matching degree between the reference multimedia resource and the target multimedia resource is greater than the matching degree between other candidate multimedia resources and the target multimedia resource.
  • a candidate multimedia resource whose matching degree with the target multimedia resource is greater than a second threshold is selected and determined as the reference multimedia resource.
  • the second threshold is a value set arbitrarily, such as 0.5 or 0.6.
  • the degree of matching between the candidate multimedia resource and the target multimedia resource is greater than the second threshold, indicating that the candidate multimedia resource matches the target multimedia resource, and the information of the candidate multimedia resource is determined as the information of the target multimedia resource.
  • the multiple candidate multimedia resources are arranged in ascending order of matching degree with the target multimedia resource, and a preset number of devices with the largest matching degree are selected from the multiple candidate multimedia resources. Select multimedia resources and determine them as reference multimedia resources.
  • the preset number is a value set arbitrarily, such as 3 or 4, etc. Since the greater the degree of matching between the candidate multimedia resources and the target multimedia resources, the higher the degree of matching between the candidate multimedia resources and the target multimedia resources. Therefore, the preset number of the most matched multimedia resources is selected from the multiple candidate multimedia resources. Alternative multimedia resources are used as reference multimedia resources.
  • the embodiment of this application is based on the degree of matching between the candidate multimedia resource and the target multimedia resource, and the reference multimedia resource is determined for description. In another embodiment, there is no need to perform steps 202-203, and other methods may be adopted. Determine the reference multimedia resource that matches the target multimedia resource.
  • the computer device obtains the degree of matching between each of the multiple pieces of information and the target multimedia resource according to multiple pieces of information of the reference multimedia resource.
  • the information of any multimedia resource is content information
  • the computer device obtains the matching degree between each content information of the multiple content information and the target multimedia resource according to multiple content information of the reference multimedia resource.
  • the content information of any multimedia resource includes comment information, detailed information, or other information, etc.
  • the comment information is information generated by commenting on the multimedia resource
  • the detailed information is information for introducing the multimedia resource.
  • the detailed information includes The resource type of the multimedia resource, the summary of the multimedia resource, etc.
  • the matching degree between any content information of the reference multimedia resource and the target multimedia resource is used to indicate the matching degree between the content information and the target multimedia resource. The greater the matching degree, the more matching the content information and the target multimedia, and the smaller the matching degree, indicating the content The more the information does not match the target multimedia. Since the reference multimedia resource has multiple content information, by determining the degree of matching between each content information and the target multimedia resource, the subsequent content information can be accurately determined for the target multimedia resource.
  • the computer device performs word segmentation processing on any information to obtain multiple second words; each second word and the target multimedia resource are obtained according to the frequency of each second word in the target multimedia resource The degree of matching; according to the weights of the plurality of second words, the degree of matching between the plurality of second words and the target multimedia resource is weighted to obtain the degree of matching between any information and the target multimedia resource.
  • the manner of obtaining the matching degree between content information and the target multimedia resource includes the following steps 2041-2043:
  • This step is similar to step 1 above, and will not be repeated here.
  • the degree of matching between the second word and the target multimedia resource indicates the degree of matching between the second word and the target multimedia resource.
  • the step 2042 includes: obtaining each second word according to the first frequency of occurrence of any second word in the target multimedia resource and the second frequency of occurrence of the second word in the content information. The matching degree between the two words and the target multimedia resource.
  • the second term for any q i, q i first words in the second appearance frequency f i in the target resource, the second term q i which appear in the second content information satisfy the following relationship:
  • q i represents the i-th second word among multiple second words
  • d represents the target multimedia resource
  • M is an adjustment parameter
  • dl is used to indicate the length of the target multimedia resource.
  • the length of the target multimedia resource refers to the number of words included in the text resource.
  • the length of the target multimedia resource refers to the number of words included in the title and profile information corresponding to the video resource or image resource; avg is used to represent the average value, and avgdl represents multiple candidate multimedia resources and the target multimedia resource The average length.
  • b is used to adjust the influence of the length of the multimedia resource on the matching degree.
  • the larger b is, the greater the influence of the length of the multimedia resource on the matching degree between the second word and the target multimedia resource is, and the smaller b is, the multimedia The length of the resource has less influence on the match between the second word and the target multimedia resource.
  • Relative length of multimedia resources The larger the value, the larger the M, and the smaller the matching degree between the second word and the target multimedia resource will be.
  • the second occurrence frequency qf i of the second word q i in the content information is 1, then the second word q i in the target resource
  • the first occurrence frequency f i and the matching degree R(q i , d) between the second word q i and the target multimedia resource d satisfy the following relationship:
  • the weight of the second word is used to indicate the degree of matching between the second word and the target multimedia resource, and the degree of contribution to the degree of matching between the content information and the target multimedia resource. The greater the weight, the match between the second word and the target multimedia resource The greater the degree of contribution.
  • this step 2043 includes: determining the product of the matching degree between each second word and the target multimedia resource and the corresponding weight, and determining the sum of the obtained multiple products as the content The degree of match between the information and the target multimedia resources.
  • the weighted summation of the degree of matching between multiple second words and the target multimedia resource is performed to obtain the content information and the target multimedia resource The degree of match.
  • the degree of matching between the multiple second words and the target multimedia resource, the weight of each second word and the second degree of matching Score(Q, d) between the content information Q and the target multimedia resource d satisfy The following relationships:
  • n represents the total number of multiple second words, n is a positive integer not less than 2
  • W i represents the weight of the i-th second word q i among multiple second words
  • R(q i , d) Indicates the degree of matching between the second word q i and the target multimedia resource d.
  • this step 2043 includes: determining the product of the matching degree between each second word and the target multimedia resource and the corresponding weight, and comparing the sum of the obtained multiple products with the multiple first words. The ratio between the numbers of the two words is determined as the second degree of matching between the content information and the target multimedia resource. By determining the degree of matching of each second word with the target multimedia resource and the weight of each second word, the matching degree of multiple second words with the target multimedia resource is weighted and averaged to obtain the content information and the target multimedia resource. suitability.
  • the first number of multimedia resources containing the second word is determined according to a plurality of multimedia resources and the second word
  • the weight of the second word is determined according to the first number and the total number of the multiple multimedia resources.
  • the multiple multimedia resources include target multimedia resources and multiple candidate multimedia resources, and may also include multiple preset multimedia resources.
  • a second term for any q i, the first n-number (q i), the total number N and the right second term weights q i W i satisfy the following relation:
  • c is the adjustment parameter, which is a constant set arbitrarily, for example, c is 0.5.
  • the larger the first number of multimedia resources containing the second word q i the greater the degree of discrimination of the second word q i in different multimedia resources, according to q i when the second term determines the matching degree of the target multimedia resources, the second lower degree of importance of words q i, q i so that the second words the lower the weight.
  • the second occurrence frequency qf i of each second word q i in the content information is 1, and the matching degree Score(Q, d) between the content information Q and the target multimedia resource d satisfies the following relation:
  • the computer device determines the degree of matching between each piece of information and the target multimedia resource according to the feature vector of each piece of information and the feature vector of the target multimedia resource.
  • this step 204 includes: determining the degree of matching between each content information and the target multimedia resource according to the feature vector of each content information and the feature vector of the target multimedia resource.
  • the feature vector of content information is a feature vector used to represent the content information, and different content information has different feature vectors.
  • This step is similar to the process of determining the degree of matching between each candidate multimedia resource and the target multimedia resource in step 202, and will not be repeated here.
  • the computer device selects the information of the target multimedia resource from the multiple pieces of information according to the degree of matching between each of the pieces of information and the target multimedia resource.
  • the computer device selects the target multimedia resource information from the multiple content information according to the degree of matching between each content information in the multiple content information and the target multimedia resource.
  • the matching degree between the content information of the target multimedia resource and the target multimedia resource is greater than the matching degree between other content information and the target multimedia resource.
  • the method of selecting the content information of the target multimedia resource in a possible implementation manner, from a plurality of content information, select the content information whose matching degree with the target multimedia resource is greater than the third threshold, and determine it as the content information of the target multimedia resource .
  • the third threshold is a value set arbitrarily, such as 0.5 or 0.6.
  • the degree of matching between the content information and the target multimedia resource is greater than the third threshold, which means that the content information is related to the target multimedia resource, and the content information is determined as the content information of the target multimedia resource.
  • the multiple content information is arranged in ascending order of matching degree with the target multimedia resource, and the reference number of content information with the largest matching degree is selected from the multiple content information and determined as the reference Multimedia resources.
  • the reference number is a value set arbitrarily, such as 3 or 4. Since the greater the matching degree between the content information and the target multimedia resource, the higher the matching degree between the content information and the target multimedia resource. Therefore, the reference number of content information with the largest matching degree is selected from the multiple content information as the target multimedia resource. Content information.
  • steps 203-204 are not required to be performed, and other methods may be adopted. , Determine the content information from multiple content information.
  • the computer device selects information that belongs to the target classification label and matches the target multimedia resource from the plurality of information and the classification label of each information, and determines the selected information as the target multimedia Resource information.
  • the classification label of the information in a possible implementation manner, multiple pieces of information are processed based on the classification model, and the classification label of each information is determined.
  • the information is content information
  • the information is content information
  • the classification label of each content information select content belonging to the target classification label and matching the target multimedia resource from the multiple content information Information, determined as content information.
  • the classification label is used to describe the category to which the content information belongs, and the classification label includes a poor-quality classification label and a high-quality classification label, and the inferior classification label is a vulgar classification label, a verbal classification label, and a low-quality classification label.
  • the target classification label is a high-quality classification label, so that high-quality content information is selected from a plurality of content information as the content information.
  • the classification model is a trained model, used to determine the category to which the content information belongs, and to generate a corresponding classification label for the content information.
  • For the training process of the classification model obtain multiple sample content information, preprocess the multiple sample content information, label the preprocessed multiple sample content information, determine the classification label of each sample content information, and pass The preprocessed multiple sample content information and the classification label of each sample content information train the classification model.
  • preprocessing includes traditional and simplified conversion, case conversion, hidden text removal, vulgar keyword cleaning, general filtering such as emotion filtering, sensitive filtering, and rule discrimination involving expressions, redundant character processing, and grammatical optimization. Ensure the accuracy of the sample content information. This rule determines that sensitive information such as mobile phone numbers and user accounts should be filtered out.
  • the classification label of each sample content information is determined through manual labeling.
  • the classification labels include inferior classification labels and high-quality classification labels.
  • the inferior classification labels are divided into multiple levels according to the degree of inferiority of the content information, and different degrees of inferiority correspond to different inferior classification labels; when determining the high-quality classification
  • tagging it is determined according to the content included in the content information and the number of likes corresponding to the content information.
  • the classification model when the classification model processes the content information, it obtains the feature vector of the content information, and classifies the content information by the feature vector of the content information, thereby determining the classification label of the content information.
  • the classification model includes a word vector acquisition sub-model and a classification sub-model.
  • the word vector acquisition sub-model is Text CNN (Text Convolutional Neural Networks, text convolutional neural network) model or other models.
  • the classification sub-model is SVM (Support Vector Machine). , Support Vector Machine) model or other models.
  • For user content information text analysis is used to interpret the user's focus of attention, main discussion topics, user's emotional tendency, and the main object of the main comment.
  • the content information is a variety of insights extended and expanded from this article. For example, there are interesting thought sparks, or direct postings, those favorite paragraphs and sentences, to show that you have carefully read these tiny and wonderful words from the writer.
  • the different content information of different users is equivalent to the direct expression of opinions and exchanges of various users in different fields, different levels, different world views, and different living environments. Content information allows people to discuss their views and share new information, and it can also attract people's attention and encourage page browsing.
  • the content information of the target multimedia resource is released by the user, or the target multimedia resource is processed based on the network model, and the content information is automatically generated for the target multimedia resource.
  • this kind of content information directly obtained by the network model In this way, the accuracy of the obtained content information is poor.
  • the method provided by the embodiment of the present application obtains the feature vector of the target multimedia resource and multiple candidate multimedia resources, and selects the reference multimedia resource whose feature vector matches the feature vector of the target multimedia resource from the multiple candidate multimedia resources, so that The obtained reference multimedia resource can be matched with the target multimedia resource.
  • the information matching the target multimedia resource is selected from the multiple information, and the information is determined as the information of the target multimedia resource.
  • the method of automatically determining the information for the target multimedia resource can ensure that the determined information matches the target multimedia resource and improve the accuracy of the information.
  • the number of information of the target multimedia resource is avoided to be zero, the cold start of the target multimedia resource is optimized, and the user can view the information of the target multimedia resource. Thereby increasing the attractiveness to users.
  • the accuracy of the obtained reference multimedia resource is improved, and the target is based on the degree of matching between multiple pieces of information and the target multimedia resource
  • the multimedia resource determines the information and improves the accuracy of the information obtained.
  • the information is determined from multiple pieces of information to ensure the quality of the information.
  • FIG. 6 is a schematic structural diagram of an information management system provided by an embodiment of the present application.
  • the information management system includes: a first terminal 601, a second terminal 602, and a server 603, and a first terminal 601 and a second terminal 601
  • the terminal 602 establishes a communication connection with the server 603 respectively.
  • the server 603 publishes multiple multimedia resources for users to view.
  • the first terminal 601 views the first multimedia resources published by the server 603, and generates first content information for the first multimedia resources.
  • the server 603 adopts the method provided in this embodiment of the application. ,
  • the first content information is migrated, the first content information is used as the content information of the second multimedia resource, and the second terminal 602 views the second multimedia resource when viewing the second multimedia resource released by the server 603
  • the first content information corresponding to the resource is provided in this embodiment of the application.
  • FIG. 7 is a schematic structural diagram of an information management system provided by an embodiment of the present application.
  • the information management system includes: a first user terminal 701, an uplink and downlink content interface server 702, a multimedia resource database 703, and a dispatch center Server 704, weight removal service subsystem 705, manual review subsystem 706, content distribution export server 707, statistical reporting interface server 708, content information database 709, content information quality evaluation server 710, multimedia resource matching server 711, content information matching server 712, a content information migration server 713, a second user terminal 714, and a third user terminal 715.
  • content information Taking information as content information as an example, the publishing process of multimedia resources, the publishing process of content information, and the migration process of content information are described below.
  • the first stage, multimedia resource release stage is a multimedia resource release stage
  • the first user terminal 701 is a multimedia resource generating terminal.
  • the multimedia resource generating terminal is PGC (Professional Generated Content, an institution or organization that professionally produces content), UGC (User Generated Content, user original content), MCN (Multi-Channel Network, Multi-channel network) or PUGC (Professional User Generated Content, professional user original content) content producers, through the terminal or back-end interface API (Application Programming Interface, application program interface) system, provide multimedia resources, such as local or captured images Text content, video or atlas content, music, filter templates and graphic beautification functions selected during the shooting process of this map text content, through communication with the upstream and downstream content interface server 702, first obtain the upload server interface address , And then upload multimedia resources to the uplink and downlink content interface server 702 according to the upload server interface address.
  • PGC Professional Generated Content, an institution or organization that professionally produces content
  • UGC User Generated Content, user original content
  • MCN Multi-Channel Network, Multi-channel network
  • the uplink and downlink content interface server 702 receives the multimedia resource uploaded by the first user terminal 701, and writes the multimedia resource and the meta information of the multimedia resource into the multimedia resource database 703.
  • the multimedia resource also includes a title, a publisher, a summary, and a cover image. , Release time, this meta-information includes information such as file size, cover image link, code rate, file format, title, release time, author, original mark, etc.
  • the uplink and downlink content interface server 702 submits the uploaded multimedia resources to the dispatch center server 704.
  • the multimedia resource database 703 is used to store multimedia resources and the meta-information of the multimedia resources. It also includes the categories and label information determined by the manual review subsystem 706 for the multimedia resources. For example, the categories include the first category, the second category, and the third category. The first category is technology, the second category is smart phones, and the third category is domestic mobile phones.
  • the tag information includes scenic spots XX, and the multimedia resources are text resources that introduce scenic spots XX.
  • the dispatching center server 704 is responsible for the entire dispatching process of multimedia resource circulation. It receives the multimedia resources in the library through the uplink and downlink content interface server 702, and then obtains the meta-information of the multimedia resources from the multimedia resource database 703, and dispatches the manual review subsystem 706 and resets
  • the service subsystem 705 processes multimedia resources and controls the order and priority of scheduling.
  • the deduplication service subsystem 705 is used to perform deduplication services on multimedia resources.
  • the multimedia resources are vectorized, and then an index of the vector is established, and then the distance between the vectors is compared to determine the degree of similarity;
  • Multimedia resources are vectorized by BERT (Bidirectional Encoder Representation from Transformers).
  • the deduplication service subsystem 705 writes the deduplication processing result into the multimedia resource database 703, and the manual review subsystem 706 will not perform repeated secondary processing for completely repeated multimedia resources.
  • the manual review subsystem 706 reads the multimedia resources in the multimedia resource database 703, and manually performs a round of preliminary filtering on whether the content involves pornography, gambling, and politically sensitive features. On the basis of the preliminary review, the secondary review of multimedia resources is to classify and label the multimedia resources or to confirm. Since the deduplication service subsystem 705 is not highly accurate in reviewing video resources, the deduplication service subsystem 705 conducts a second manual review process on the basis of the video resource review, thereby improving the labeling of video resources. Accuracy, and improve processing efficiency.
  • the dispatch center server 704 also sends the multimedia resources passed through the manual review subsystem 706 to the content distribution export server 707.
  • the content distribution export server 707 receives the multimedia resources that have passed the manual review subsystem 706 sent by the dispatch center server 704, and publishes the multimedia resources in the form of Feeds for viewing by the second user terminal 714.
  • the second stage, content information release stage is the second stage, content information release stage
  • the third user terminal 715 obtains index information for accessing multimedia resources through communication with the uplink and downlink content interface server 702, and then downloads the corresponding multimedia resources and plays them through a local player.
  • the behavior data, freezes, loading time, playback clicks, etc. during the upload and download process are reported to the statistical reporting interface server 708.
  • the interactive information of the third user terminal 715 on the multimedia resources such as comments, likes, forwarding, and collection of the multimedia resources, is reported to the statistical reporting interface server 708.
  • the third user terminal 715 can also report low-quality content information in the multimedia resources, report the reported content information to the content information database 709, and go through the manual review subsystem 706 before serving as a sample.
  • the statistical reporting interface server 708 receives the content information of the multimedia resource uploaded by the third user terminal 715, and writes the content information into the content information database 709.
  • the content information database 709 is used to store content information, and provide content information and other interactive data for the content information quality evaluation server 710 and the content information migration server 713.
  • the third stage, content information migration stage is the third stage, content information migration stage.
  • the content information quality evaluation server 710 performs quality modeling and classification on the content information of the multimedia resources according to the method provided in the foregoing embodiment, so as to obtain the classification label of each content information.
  • the multimedia resource matching server 711 determines the matched reference multimedia resource for the target multimedia resource according to the method provided in the foregoing embodiment, and provides the content information migration server 713 with the matching result.
  • the content information matching server 712 determines the content information matching the target multimedia resource among the multiple content information of the reference multimedia resource according to the method provided in the foregoing embodiment, and provides the content information migration server 713 with the matching result.
  • the content information migration server 713 realizes the effect of content information migration and generates content information for the target multimedia resources according to the matching results of the multimedia resource matching server 711 and the content information matching server 712 through the scheduling server.
  • the dispatch center server 704 also schedules content information migration services to complete the migration of high-quality content information, and at the same time outputs the migrated content information to the content distribution export server 707, which sends the content distribution export server 707 to the second user terminal 714 for the second user terminal 714. Second, the user terminal 714 checks.
  • FIG. 8 is a schematic structural diagram of an information determination device provided by an embodiment of the present application. As shown in FIG. 8, the device includes:
  • the feature vector obtaining module 801 is used to obtain feature vectors of the target multimedia resource and feature vectors of multiple candidate multimedia resources;
  • the multimedia resource selection module 802 is configured to select, from a plurality of candidate multimedia resources, a reference multimedia resource whose feature vector matches the feature vector of the target multimedia resource;
  • the information determining module 803 is configured to select information matching the target multimedia resource from the multiple pieces of information according to multiple pieces of information of the reference multimedia resource, and determine the selected information as the information of the target multimedia resource.
  • the device provided by the embodiment of the present application obtains the feature vector of the target multimedia resource and multiple candidate multimedia resources, and selects the reference multimedia resource whose feature vector matches the feature vector of the target multimedia resource from the multiple candidate multimedia resources, so that The obtained reference multimedia resource can be matched with the target multimedia resource.
  • the information matching the target multimedia resource is selected from the multiple information, and the information is determined as the information of the target multimedia resource.
  • the method of automatically determining the information for the target multimedia resource can ensure that the determined information matches the target multimedia resource and improve the accuracy of the information.
  • the feature vector obtaining module 801 includes:
  • the graph network creation unit 8011 is used to create a graph network according to the target multimedia resource and multiple candidate multimedia resources.
  • the graph network includes the target multimedia resource node corresponding to the target multimedia resource and multiple candidate multimedia resources corresponding to the multiple candidate multimedia resources. Resource node, any two multimedia resource nodes meeting the first association condition are connected;
  • the first feature vector obtaining unit 8012 is configured to process the graph network based on the first feature extraction model, obtain the feature vector of the target multimedia resource node and the feature vector of multiple candidate multimedia resource nodes, and convert the feature vector of the target multimedia resource node As the feature vector of the target multimedia resource, the feature vector of multiple candidate multimedia resource nodes is used as the feature vector of multiple candidate multimedia resources.
  • the graph network creation unit 8011 is also used to perform word segmentation processing on the text resources in the target multimedia resource and the text resources in the multiple candidate multimedia resources to obtain multiple first words; according to the target multimedia resources, multiple The candidate multimedia resources and multiple first words are created to create a graph network.
  • the graph network includes target multimedia resource nodes corresponding to the target multimedia resources, multiple candidate multimedia resource nodes corresponding to multiple candidate multimedia resources, and multiple first word correspondences The multiple word nodes of, and the word nodes meeting the second association condition are connected to the multimedia resource node.
  • the device further includes:
  • the node determination module 804 is configured to determine that the word node corresponding to the first word and the multimedia resource node corresponding to the multimedia resource satisfy the second association in response to the occurrence frequency of any first word in the text resource of any multimedia resource being greater than the first threshold. condition.
  • the multimedia resources include video resources;
  • the feature vector acquisition module 801 includes:
  • the frame extraction processing unit 8013 is configured to perform frame extraction processing on the target multimedia resource and multiple candidate multimedia resources respectively to obtain multiple video frames corresponding to the target multimedia resource and multiple video frames corresponding to the multiple candidate multimedia resources;
  • the second feature vector acquiring unit 8014 is configured to process multiple video frames corresponding to the target multimedia resource and multiple video frames corresponding to multiple candidate multimedia resources, respectively, based on the second feature extraction model, to obtain the characteristics of the target multimedia resource Vectors and feature vectors of multiple candidate multimedia resources.
  • the second feature vector acquiring unit 8014 is further configured to process multiple video frames corresponding to any multimedia resource separately based on the second feature extraction model to obtain feature vectors of multiple video frames; The feature vector of the frame is merged to obtain the feature vector of the multimedia resource.
  • the multimedia resource selection module 802 includes:
  • the first matching degree obtaining unit 8021 is configured to obtain the matching degree between each candidate multimedia resource and the target multimedia resource according to the feature vector of the target multimedia resource and the feature vector of multiple candidate multimedia resources;
  • the multimedia resource selection unit 8022 is used to select a reference multimedia resource from a plurality of candidate multimedia resources according to the degree of matching between each candidate multimedia resource and the target multimedia resource, and the degree of matching between the reference multimedia resource and the target multimedia resource is greater than that of other candidates The degree of matching between the multimedia resources and the target multimedia resources.
  • the information determining module 803 includes:
  • the second matching degree obtaining unit 8031 is configured to obtain the matching degree between each of the multiple pieces of information and the target multimedia resource according to multiple pieces of information of the reference multimedia resource;
  • the information selection unit 8032 is used to select the information of the target multimedia resource from multiple pieces of information according to the degree of matching between each information and the target multimedia resource, and the degree of matching between the information of the target multimedia resource and the target multimedia resource is greater than that of other information and the target multimedia resource The degree of match.
  • the second matching degree obtaining unit 8031 is further configured to perform word segmentation processing on any information to obtain multiple second words; according to the frequency of each second word in the target multimedia resource, obtain each second word The matching degree between the words and the target multimedia resources; according to the weights of the multiple second words, the matching degrees between the multiple second words and the target multimedia resources are weighted to obtain the matching degree between the information and the target multimedia resources.
  • the second matching degree obtaining unit 8031 is further configured to determine the degree of matching between each piece of information and the target multimedia resource according to the feature vector of each piece of information and the feature vector of the target multimedia resource.
  • the information determining module 803 includes:
  • the information determining unit 8033 is configured to select information belonging to the target classification label and matching the target multimedia resource from the plurality of information according to the plurality of information and the classification label of each information, and determine the selected information as the information of the target multimedia resource .
  • the device further includes:
  • the classification label determination module 805 is configured to process multiple pieces of information based on the classification model, and determine the classification label of each piece of information.
  • FIG. 10 is a schematic structural diagram of a terminal provided by an embodiment of the present application, which implements the operations performed by the computer device in the foregoing embodiment.
  • the terminal 1000 is a portable mobile terminal, such as: smart phones, tablet computers, MP3 players (Moving Picture Experts Group Audio Layer III, dynamic image expert compression standard audio layer 3), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic image Experts compress the standard audio level 4) Players, laptops, desktop computers, head-mounted devices, smart TVs, smart speakers, smart remotes, smart microphones, or any other smart terminals.
  • the terminal 1000 may also be called user equipment, portable terminal, laptop terminal, desktop terminal and other names.
  • the terminal 1000 includes a processor 1001 and a memory 1002.
  • the processor 1001 includes one or more processing cores, such as a 4-core processor, an 8-core processor, and so on.
  • the memory 1002 includes one or more computer-readable storage media.
  • the computer-readable storage media is non-transitory and used to store at least one instruction.
  • the at least one instruction is used by the processor 1001 to implement the method in this application. How to determine the information provided by the example.
  • the terminal 1000 optionally further includes: a peripheral device interface 1003 and at least one peripheral device.
  • the processor 1001, the memory 1002, and the peripheral device interface 1003 are connected by a bus or signal line.
  • Each peripheral device is connected to the peripheral device interface 1003 through a bus, a signal line or a circuit board.
  • the peripheral device includes: at least one of a radio frequency circuit 1004, a display screen 1005, and an audio circuit 1006.
  • the radio frequency circuit 1004 is used for receiving and transmitting RF (Radio Frequency, radio frequency) signals, also called electromagnetic signals.
  • the radio frequency circuit 1004 communicates with a communication network and other communication devices through electromagnetic signals.
  • the display screen 1005 is used to display a UI (User Interface, user interface).
  • the UI includes graphics, text, icons, videos, and any combination of them.
  • the display screen 1005 is a touch display screen, and is also used to provide virtual buttons and/or virtual keyboards.
  • the audio circuit 1006 includes a microphone and a speaker.
  • the microphone is used to collect audio signals of the user and the environment, and convert the audio signals into electrical signals and input to the processor 1001 for processing, or input to the radio frequency circuit 1004 to implement voice communication.
  • the microphone is also an array microphone or an omnidirectional acquisition microphone.
  • the speaker is used to convert the electrical signal from the processor 1001 or the radio frequency circuit 1004 into an audio signal.
  • FIG. 10 does not constitute a limitation on the terminal 1000, and includes more or fewer components than shown, or some components are combined, or different component arrangements are adopted.
  • FIG. 11 is a schematic structural diagram of a server provided by an embodiment of the present application.
  • the server 1100 may have relatively large differences due to different configurations or performance, including one or more processors (Central Processing Units, CPU) 1101 and one or More than one memory 1102, where at least one instruction is stored in the memory 1102, and the at least one instruction is loaded and executed by the processor 1101 to implement the methods provided in the foregoing method embodiments.
  • processors Central Processing Units, CPU
  • the server also has components such as a wired or wireless network interface, a keyboard, an input and output interface for input and output, and the server also includes other components for realizing device functions, which will not be repeated here.
  • the server 1100 is configured to execute the above-mentioned information determination method.
  • An embodiment of the present application also provides a computer device, which includes a processor and a memory, and at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement the information determination method of the foregoing embodiment.
  • An embodiment of the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor, so as to implement the information determination method of the foregoing embodiment.
  • An embodiment of the present application also provides a computer program, in which at least one instruction is stored, and the at least one instruction is loaded and executed by a processor, so as to implement the information determination method of the foregoing embodiment.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请实施例公开了一种信息确定方法、装置、计算机设备及存储介质,属于计算机技术领域。该方法包括:获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量,从多个备选多媒体资源中,选取特征向量与目标多媒体资源的特征向量匹配的参考多媒体资源,以使获取到的参考多媒体资源能够与该目标多媒体资源相匹配,根据参考多媒体资源的多个信息,从多个信息中选取与目标多媒体资源匹配的信息,将选取的信息确定为目标多媒体资源的信息,提供了一种自动为目标多媒体资源确定信息的方式,能够保证确定信息与该目标多媒体资源相匹配,提高了信息的准确性。

Description

信息确定方法、装置、计算机设备及存储介质
本申请要求于2020年03月24日提交、申请号为202010213262.1、发明名称为“内容信息确定方法、装置、计算机设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及计算机技术领域,特别涉及一种信息确定方法、装置、计算机设备及存储介质。
背景技术
随着互联网的快速发展和广泛普及,互联网中传播的多媒体资源越来越丰富。通常用户在观看多媒体资源时,还能够针对多媒体资源发布信息,其他用户在查看该多媒体资源时,能够查看到该用户发布的信息。
发明内容
本申请实施例提供了一种信息确定方法、装置、计算机设备及存储介质,能够提高内容信息确定的准确性。所述技术方案如下:
一方面,提供了一种信息确定方法,所述方法包括:
获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量;
从所述多个备选多媒体资源中,选取特征向量与所述目标多媒体资源的特征向量匹配的参考多媒体资源;
根据所述参考多媒体资源的多个信息,从所述多个信息中选取与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息。
另一方面,提供了一种信息确定装置,所述装置包括:
特征向量获取模块,用于获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量;
多媒体资源选取模块,用于从所述多个备选多媒体资源中,选取特征向量与所述目标多媒体资源的特征向量匹配的参考多媒体资源;
信息确定模块,用于根据所述参考多媒体资源的多个信息,从所述多个信息中选取与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息。
可选地,所述特征向量获取模块,包括:
图网络创建单元,用于根据所述目标多媒体资源及所述多个备选多媒体资源,创建图网络,所述图网络包括所述目标多媒体资源对应的目标多媒体资源节点及所述多个备选多媒体资源对应的多个备选多媒体资源节点,满足第一关联条件的任两个多媒体资源节点连接;
第一特征向量获取单元,用于基于第一特征提取模型对所述图网络进行处理,获取所述目标多媒体资源节点的特征向量及所述多个备选多媒体资源节点的特征向量,将所述目标多媒体资源节点的特征向量作为所述目标多媒体资源的特征向量,将所述多个备选多媒体资源节点的特征向量作为所述多个备选多媒体资源的特征向量。
可选地,所述图网络创建单元,还用于对所述目标多媒体资源中的文本资源及所述多个备选多媒体资源中的文本资源进行分词处理,得到多个第一词语;根据所述目标多媒体资源、所述多个备选多媒体资源及所述多个第一词语,创建图网络,所述图网络包括所述目标多媒体资源对应的目标多媒体资源节点、所述多个备选多媒体资源对应的多个备选多媒体资源节点及所述多个第一词语对应的多个词节点,满足第二关联条件的词节点与多媒体资源节点连接。
可选地,所述装置还包括:
节点确定模块,用于响应于任一第一词语在任一多媒体资源的文本资源中的出现频率大于第一阈值,确定所述第一词语对应的词节点与所述多媒体资源对应的多媒体资源节点满足所述第二关联条件。
可选地,多媒体资源包括视频资源;所述特征向量获取模块,包括:
抽帧处理单元,用于分别对所述目标多媒体资源及所述多个备选多媒体资源进行抽帧处理,得到所述目标多媒体资源对应的多个视频帧及所述多个备选多媒体资源对应的多个视频帧;
第二特征向量获取单元,用于基于第二特征提取模型,分别对所述目标多 媒体资源对应的多个视频帧及所述多个备选多媒体资源对应的多个视频帧进行处理,得到所述目标多媒体资源的特征向量及所述多个备选多媒体资源的特征向量。
可选地,所述第二特征向量获取单元,还用于基于所述第二特征提取模型,对任一多媒体资源对应的多个视频帧分别进行处理,得到所述多个视频帧的特征向量;对所述多个视频帧的特征向量进行融合,得到所述多媒体资源的特征向量。
可选地,所述多媒体资源选取模块,包括:
第一匹配度获取单元,用于根据所述目标多媒体资源的特征向量及所述多个备选多媒体资源的特征向量,获取每个备选多媒体资源与所述目标多媒体资源的匹配度;
多媒体资源选取单元,用于根据所述每个备选多媒体资源与所述目标多媒体资源的匹配度,从所述多个备选多媒体资源中选取参考多媒体资源,所述参考多媒体资源与所述目标多媒体资源的匹配度大于其他备选多媒体资源与所述目标多媒体资源的匹配度。
可选地,所述信息确定模块,包括:
第二匹配度获取单元,用于根据所述参考多媒体资源的多个信息,获取所述多个信息中每个信息与所述目标多媒体资源的匹配度;
信息选取单元,用于根据所述每个信息与所述目标多媒体资源的匹配度,从所述多个信息中选取所述目标多媒体资源的信息,所述目标多媒体资源的信息与所述目标多媒体资源的匹配度大于其他信息与所述目标多媒体资源的匹配度。
可选地,所述第二匹配度获取单元,还用于对任一信息进行分词处理,得到多个第二词语;根据每个第二词语在所述目标多媒体资源中的出现频率,获取所述每个第二词语与所述目标多媒体资源的匹配度;根据所述多个第二词语的权重,对所述多个第二词语与所述目标多媒体资源的匹配度进行加权,得到所述信息与所述目标多媒体资源的匹配度。
可选地,所述第二匹配度获取单元,还用于根据所述每个信息的特征向量及所述目标多媒体资源的特征向量,确定所述每个信息与所述目标多媒体资源的匹配度。
可选地,所述信息确定模块,包括:
信息确定单元,用于根据所述多个信息及每个信息的分类标签,从所述多个信息中选取属于目标分类标签、且与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息。
可选地,所述装置还包括:
分类标签确定模块,用于基于分类模型对所述多个信息进行处理,确定所述每个信息的分类标签。
另一方面,提供了一种计算机设备,所述计算机设备包括处理器和存储器,所述存储器中存储有至少一条指令,所述至少一条指令由所述处理器加载并执行,以实现如上述方面所述的信息确定方法。
另一方面,提供了一种计算机可读存储介质,所述计算机可读存储介质中存储有至少一条指令,所述至少一条指令由处理器加载并执行,以实现如上述方面所述的信息确定方法。
本申请实施例提供的技术方案带来的有益效果至少包括:
本申请实施例提供的方法、装置、计算机设备及存储介质,获取目标多媒体资源及多个备选多媒体资源的特征向量,从多个备选多媒体资源中,选取特征向量与目标多媒体资源的特征向量匹配的参考多媒体资源,以使获取到的参考多媒体资源能够与该目标多媒体资源相匹配,根据参考多媒体资源的多个信息,从多个信息中选取与目标多媒体资源匹配的信息,将选取的信息确定为目标多媒体资源的信息,提供了一种自动为目标多媒体资源确定信息的方式,能够保证确定信息与该目标多媒体资源相匹配,提高了信息的准确性。
附图说明
为了更清楚地说明本申请实施例中的技术方案,下面将对实施例描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请实施例的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1是本申请实施例提供的一种实施环境的示意图;
图2是本申请实施例提供的一种信息确定方法的流程图;
图3是本申请实施例提供的一种信息确定方法的流程图;
图4是本申请实施例提供的一种信息确定方法的流程图;
图5是本申请实施例提供的一种信息确定方法的流程图;
图6是本申请实施例提供的一种信息管理系统的结构示意图;
图7是本申请实施例提供的一种信息管理系统的结构示意图;
图8是本申请实施例提供的一种信息确定装置的结构示意图;
图9是本申请实施例提供的一种信息确定装置的结构示意图;
图10是本申请实施例提供的一种终端的结构示意图;
图11是本申请实施例提供的一种服务器的结构示意图。
具体实施方式
为使本申请实施例的目的、技术方案和优点更加清楚,下面将结合附图对本申请实施方式作进一步地详细描述。
本申请所使用的术语“第一”、“第二”、“第三”等可在本文中用于描述各种概念,但除非特别说明,这些概念不受这些术语限制。这些术语仅用于将一个概念与另一个概念区分。举例来说,在不脱离本申请的范围的情况下,可以将第一阈值称为第二阈值,且类似地,可将第二阈值称为第一阈值。
本申请所使用的术语“至少一个”、“多个”、“每个”、“任一”,至少一个包括一个、两个或两个以上,多个包括两个或两个以上,而每个是指对应的多个中的每一个,任一是指多个中的任意一个。举例来说,多个元素包括3个元素,而每个是指这3个元素中的每一个元素,任一是指这3个元素中的任意一个,可以是第一个,可以是第二个、也可以是第三个。
为了便于理解本申请实施例的技术过程,下面对本申请实施例所涉及的一些名词进行解释:
Feeds(一种信息形式):网站以Feeds的形式将信息传播给用户,以时间轴方式排列,Timeline(时间线)是Feeds最原始、最直觉也最基本的展示形式。
人工智能(Artificial Intelligence,AI)是利用数字计算机或者数字计算机控制 的机器模拟、延伸和扩展人的智能,感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。换句话说,人工智能是计算机科学的一个综合技术,它企图了解智能的实质,并生产出一种新的能以人类智能相似的方式做出反应的智能机器。人工智能也就是研究各种智能机器的设计原理与实现方法,使机器具有感知、推理与决策的功能。
人工智能技术是一门综合学科,涉及领域广泛,既有硬件层面的技术也有软件层面的技术。人工智能基础技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理技术、操作/交互系统、机电一体化等技术。人工智能软件技术主要包括计算机视觉技术、语音处理技术、自然语言处理技术以及机器学习/深度学习等几大方向。
自然语言处理(Nature Language processing,NLP)是计算机科学领域与人工智能领域中的一个重要方向。它研究能实现人与计算机之间用自然语言进行有效通信的各种理论和方法。自然语言处理是一门融语言学、计算机科学、数学于一体的科学。因此,这一领域的研究将涉及自然语言,即人们日常使用的语言,所以它与语言学的研究有着密切的联系。自然语言处理技术通常包括文本处理、语义理解、机器翻译、机器人问答、知识图谱等技术。
机器学习(Machine Learning,ML)是一门多领域交叉学科,涉及概率论、统计学、逼近论、凸分析、算法复杂度理论等多门学科。专门研究计算机怎样模拟或实现人类的学习行为,以获取新的知识或技能,重新组织已有的知识结构使之不断改善自身的性能。机器学习是人工智能的核心,是使计算机具有智能的根本途径,其应用遍及人工智能的各个领域。机器学习和深度学习通常包括人工神经网络、置信网络、强化学习、迁移学习、归纳学习、示教学习等技术。
深度学习:深度学习的概念源于人工神经网络的研究,包含多个隐含层的多层感知器就是一种深度学习结构。深度学习通过组合低层特征,形成更加抽象的高层表示属性类别或特征,以发现数据的分布式特征表示。
本申请实施例提供的方案,基于人工智能的机器学习技术,训练第一特征提取模型和第二特征提取模型,利用训练后的第一特征提取模型和第二特征提取模型,获取多媒体资源的特征向量,后续实现为目标多媒体资源确定内容信息。
图1是本申请实施例提供的一种实施环境的示意图,如图1所示,该实施环境包括终端101和服务器102。终端101与服务器102建立通信连接,通过建立的通信连接进行交互。其中,该终端101为手机、计算机、平板电脑等多种类型的终端。服务器102为一台服务器,或者是由若干台服务器组成的服务器集群,或者是一个云计算服务器中心。
服务器102根据目标多媒体资源的特征向量及多个备选多媒体资源的特征向量,确定特征向量与目标多媒体资源的特征向量匹配的参考多媒体资源,从参考多媒体资源到的多个信息中选取与目标多媒体资源匹配的信息,确定为目标多媒体资源的信息,终端101显示目标多媒体资源时,服务器102向该终端101发送目标多媒体资源及对应的信息,供用户查看。
本申请实施例提供的方法,可用于为多媒体资源确定内容信息的场景。
例如,视频资源的评论信息迁移场景下:
计算机设备在获取到视频资源后,采用本申请实施例提供的评论信息确定方法,将其他视频资源的评论信息进行迁移,作为该视频资源的评论信息,以使其他用户在观看该视频资源时,能够查看到该评论信息,从而提高了该视频资源的评论信息数量,进而提高了该视频资源的热门度。
再例如,文章的评论信息迁移场景下:
应用服务器在获取到用户发布的文章后,采用本申请实施例提供的评论信息确定方法,将其他文章的多个评论信息进行迁移,作为该文章的评论信息,以使新发布的文章具有多个评论信息,从而提高了该文章的评论信息数量。用户通过查看各个文章的评论信息的数量,选择该文章进行查看,提高了该文章对用户的吸引力。
本申请实施例提供的信息确定方法,应用于计算机设备中,该计算机设备包括终端或服务器,该终端为手机、计算机、平板电脑等多种类型的终端,该服务器为一台服务器,或者是由若干台服务器组成的服务器集群,或者是一个云计算服务器中心。
图2是本申请实施例提供的一种信息确定方法的流程图,应用于计算机设备中,如图2所示,该方法包括:
201、计算机设备获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量。
其中,多媒体资源包括视频资源、音频资源、文本资源、图像资源等。该多媒体资源是任意发布者发布的资源,例如,是应用的管理人员发布的多媒体资源,或者是应用的用户发布的多媒体资源。
在一种可能实现方式中,终端中安装有目标应用,发布者通过该目标应用发布多媒体资源,终端将发布的多媒体资源上传至目标应用对应的应用服务器,在应用服务器中进行存储。其中,目标应用为视频应用、音乐应用、阅读应用等。
在本申请实施例中,目标多媒体资源和备选多媒体资源属于相同类型的多媒体资源,如目标多媒体资源和备选多媒体资源均为视频资源或者文本资源;或者目标多媒体资源和备选多媒体资源属于不同类型的多媒体资源,如目标多媒体资源为文本资源,备选多媒体资源为视频资源。在多个多媒体资源中,将任一多媒体资源作为目标多媒体资源,将其他的多媒体资源作为备选多媒体资源,或者将满足条件的多媒体资源作为目标多媒体资源,将其他的多媒体资源作为备选多媒体资源。例如,根据多媒体资源的信息的数量,将信息的数量小于第四阈值的多媒体资源作为目标多媒体资源,将信息的数量不小于第四阈值的多媒体资源作为备选多媒体资源。
特征向量是用于表示多媒体资源的特征信息的向量,该特征向量包括多个维度。由于不同的多媒体资源的特征信息不同,则不同的多媒体资源对应的特征向量不同。
在本申请实施例中采用以下两种方式,获取多媒体资源的特征向量。
如图3所示,第一种方式包括以下步骤2011-2012:
2011、根据目标多媒体资源及多个备选多媒体资源,创建图网络。
其中,图网络是多个节点之间连接关系的表示形式。图网络包括目标多媒体资源对应的目标多媒体资源节点及多个备选多媒体资源对应的多个备选多媒体资源节点。在多个多媒体资源节点中,将满足第一关联条件的任两个多媒体资源节点连接,得到该图网络。
对于确定多媒体资源节点是否满足第一关联条件的方式,在一种可能实现方式中,根据任两个多媒体资源的匹配度,响应于该匹配度大于第五阈值,确 定该任两个多媒体资源对应的多媒体资源节点满足第一关联条件。其中,任两个多媒体资源的匹配度用于表示这两个多媒体资源之间的匹配程度。
在一种可能实现方式中,该步骤2011包括以下步骤1-2:
步骤1、对目标多媒体资源中的文本资源及多个备选多媒体资源中的文本资源进行分词处理,得到多个第一词语。
其中,文本资源是包括文字的资源。当多媒体资源为文本资源时,该多媒体资源的文本资源也即是该多媒体资源自身;当多媒体资源为视频资源时,该多媒体资源的文本资源包括该多媒体资源的标题及简介信息,或者包括该视频资源中的字幕信息;当多媒体资源为图像资源时,该多媒体资源的文本资源包括该多媒体资源的标题及简介信息,或者包括图像资源中的文字信息;当多媒体资源为音频资源时,该多媒体资源的文本资源包括该多媒体资源的标题及简介信息,或者包括由该音频资源转换后的文本资源。
分词处理是将文本资源中连续的多个字分成词语的过程。对目标多媒体资源中的文本资源进行分词处理,并对每个备选多媒体资源中的文本资源进行分词处理,从而得到多个文本资源分词处理后的多个第一词语。
由于不同的多媒体资源中可能会包括相同的词语,因此,得到的多个第一词语中包括相同的词语。或者,在一种可能实现方式中,该步骤2011还包括:分别对多个备选多媒体资源及目标多媒体资源进行分词处理,得到多个第三词语,对该多个第三词语进行去重处理,得到该多个第一词语。通过对多个第三词语进行去重处理,将多个第三词语中相同的单词进行筛除,以使得到的多个第一词语均不同,避免了词语的重复。
步骤2、根据目标多媒体资源、多个备选多媒体资源及多个第一词语,创建图网络。
其中,图网络包括目标多媒体资源对应的目标多媒体资源节点、多个备选多媒体资源对应的多个备选多媒体资源节点及多个第一词语对应的多个词节点。在创建多个词节点时,如果多个第一词语中包括相同的第一词语,在图网络中只创建相同的第一词语对应的一个词节点。在多个词节点和多个多媒体资源节点中,将满足第二关联条件的词节点与多媒体资源节点连接。
对于确定词节点与多媒体资源节点是否满足第二关联条件的方式,在一种可能实现方式中,对于任一第一词语及任一多媒体资源,响应于该多媒体资源 的文本资源中包含该第一词语,确定该第一词语对应的词节点与该多媒体资源对应的多媒体资源节点满足第二关联条件。
在另一种可能实现方式中,响应于任一第一词语在任一多媒体资源的文本资源中的出现频率大于第一阈值,确定第一词语对应的词节点与多媒体资源对应的多媒体资源节点满足第二关联条件。
其中,出现频率表示词语在多媒体资源中的出现次数。第一阈值是任意设置的数值,如5或6等。词语在多媒体资源中的出现频率越高,则表示该词语与该多媒体资源的匹配度越高,词语在多媒体资源中的出现频率越低,则表示该词语与该多媒体资源的匹配度越低。
在构建图网络时,确定每个第一词语在每个多媒体资源中的出现频率,当任一词语在任一多媒体资源中的出现频率大于第一阈值,则将该词语对应的词节点与该多媒体资源对应的多媒体资源节点连接,从而得到该图网络。
对于确定第一词语与多媒体资源的出现频率的方式,在一种可能实现方式中,根据任一第一词语在任一多媒体资源中进行遍历,得到该第一词语在该多媒体资源中的出现次数,作为该第一词语在该多媒体资源中的出现频率。
另外,由于在多个第一词语中,不同的第一词语具有相同的含义,即不同的第一词语为同义词语。因此,在构建图网络时,通过将具有相同含义的词节点连接,使图网络中包括词节点之间的关系,从而提高了得到图网络的准确性。
2012、基于第一特征提取模型对图网络进行处理,获取目标多媒体资源节点的特征向量及多个备选多媒体资源节点的特征向量,将目标多媒体资源节点的特征向量作为目标多媒体资源的特征向量,将多个备选多媒体资源节点的特征向量作为多个备选多媒体资源的特征向量。
基于第一特征提取模型对图网络进行处理,该第一特征提取模型输出每个多媒体资源节点的特征向量,从而得到目标多媒体资源节点的特征向量及多个备选多媒体资源节点的特征向量。
第一特征提取模型为GCN(Graph Convolutional Network,图卷积网络)模型。其中,GCN模型通过对图网络中的各个节点的特征进行提取,通过Graph Embedding(图的嵌入表示),得到每个节点的特征向量,基于该GCN模型对图网络进行处理,实现图网络中的Node Classification(节点分类)、Graph Classification(图分类)、Link Prediction(边预测)。该GCN模型通过一种半监 督学习方法进行训练得到,基于训练后的GCN模型获取图网络中每个节点的特征向量;或者对该GCN模型不进行训练,基于初始化参数的GCN模型对图网络进行处理,得到每个节点的特征向量。该GCN模型获取为DeepWalk(一种神经网络)模型,或者为Word2vec(Word To Vector,一种词向量模型)。或者,第一特征提取模型为GAT(Graph Attention Network,图注意力网络)模型,该GAT模型在对图网络进行处理时,引入了Attention(注意力)机制,根据相邻节点之间的权重,确定相邻节点之间的关系,从而提高了取目标多媒体资源节点及多个备选多媒体资源节点的特征向量的准确性。
在一种可能实现方式中,基于训练后的GCN模型获取到图网络中每个节点的特征向量后,还能够将该图网络作为样本,继续对GCN模型进行训练。
在本申请实施例中,由于图网络中包括多个多媒体资源及多个词语之间的关系,后续通过图网络获取每个多媒体资源的特征向量,能够在特征向量中融入多个多媒体资源及多个词语之间的关系,从而提高了获取到的多媒体资源的特征向量的准确性。
如图4所示,第二种方式,包括以下步骤2013-2014:
2013、分别对目标多媒体资源及多个备选多媒体资源进行抽帧处理,得到目标多媒体资源对应的多个视频帧及多个备选多媒体资源对应的多个视频帧。
其中,多媒体资源包括视频资源,每个视频资源包括多个视频帧,因此需要对每个视频资源进行抽帧处理,得到每个多媒体资源对应的多个视频帧。例如,视频资源的时长为20秒,每秒一个视频帧,则对该视频资源进行抽帧处理,得到20个视频帧,也即是得到20个图像。
2014、基于第二特征提取模型,分别对目标多媒体资源对应的多个视频帧及多个备选多媒体资源对应的多个视频帧进行处理,获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量。
其中,该第二特征提取模型为用于获取多媒体资源的特征向量的模型,为CNN(Convolutional Neural Networks,卷积神经网络)模型或者其他模型。
在一种可能实现方式中,该步骤2014包括:基于第二特征提取模型,对任一多媒体资源对应的多个视频帧分别进行处理,得到多个视频帧的特征向量,对多个视频帧的特征向量进行融合,得到多媒体资源的特征向量。
对于获取视频帧的特征向量的方式,在一种可能实现方式中,该第二特征 提取模型包括特征检测子模型和全局特征聚合子模型;在获取任一视频帧的特征向量时,基于该特征检测子模型,对该视频帧的局部特征点进行检测,得到该视频帧中多个局部特征的检测结果,基于该全局特征聚合子模型对该多个局部特征进行聚合处理,得到该视频帧的特征向量。
202、根据目标多媒体资源及多个备选多媒体资源的特征向量,确定每个备选多媒体资源与目标多媒体资源的匹配度。
其中,匹配度用于表示备选多媒体资源与目标多媒体资源之间的匹配程度,匹配度越大,表示备选多媒体资源与目标多媒体资源越相关,匹配度越小,表示备选多媒体资源与目标多媒体资源越不相关。该备选多媒体资源与目标多媒体资源的匹配度,采用备选多媒体资源与目标多媒体资源的特征向量之间的余弦距离、欧几里德距离、马氏距离、切比雪夫距离、汉明距离、标准化欧氏距离等来表示。
在一种可能实现方式中,根据备选多媒体资源的特征向量A及目标多媒体资源的特征向量B,确定该备选多媒体资源与目标多媒体资源之间的匹配度similarity(A,B),该特征向量A、该特征向量B及该匹配度similarity(A,B)满足以下关系:
Figure PCTCN2021075270-appb-000001
其中,||A||用于表示特征向量A的长度,||B||用于表示特征向量B的长度,A i表示特征向量A中第i个维度的值;B i表示特征向量B中第i个维度的值,i为不小于1、且不大于n的正整数,n用于表示特征向量包含的维度个数,n为大于等于2的正整数。
203、计算机设备根据每个备选多媒体资源与目标多媒体资源的匹配度,从多个备选多媒体资源中选取参考多媒体资源。
其中,参考多媒体资源与目标多媒体资源的匹配度大于其他备选多媒体资源与目标多媒体资源的匹配度。
对于选取参考多媒体资源的方式,在一种可能实现方式中,从多个备选多媒体资源中,选取与目标多媒体资源的匹配度大于第二阈值的备选多媒体资源,确定为参考多媒体资源。
其中,第二阈值为任意设置的数值,如0.5或0.6等。备选多媒体资源与目标多媒体资源的匹配度大于第二阈值,则表示该备选多媒体资源与该目标多媒体资源匹配,则将该备选多媒体资源的信息确定为目标多媒体资源的信息。
在另一种可能实现方式中,将多个备选多媒体资源按照与目标多媒体资源的匹配度由小到大的顺序排列,从多个备选多媒体资源中选取匹配度最大的预设数目个备选多媒体资源,确定为参考多媒体资源。
其中,预设数目为任意设置的数值,如3或4等。由于备选多媒体资源与目标多媒体资源的匹配度越大,则表示备选多媒体资源与目标多媒体资源的匹配程度越高,因此,从多个备选多媒体资源中选取匹配度最大的预设数目个备选多媒体资源作为参考多媒体资源。
需要说明的是,本申请实施例是以备选多媒体资源与目标多媒体资源的匹配度,确定参考多媒体资源进行说明的,而在另一实施例中,无需执行步骤202-203,可以采取其他方式确定与目标多媒体资源匹配的参考多媒体资源。
204、计算机设备根据参考多媒体资源的多个信息,获取多个信息中每个信息与目标多媒体资源的匹配度。
在一种可能实现方式中,任一多媒体资源的信息为内容信息,计算机设备根据参考多媒体资源的多个内容信息,获取多个内容信息中每个内容信息与目标多媒体资源的匹配度。
其中,任一多媒体资源的内容信息包括评论信息、详情信息或者其他信息等,该评论信息为对多媒体资源进行评论生成的信息,该详情信息为对多媒体资源进行介绍的信息,如该详情信息包括多媒体资源的资源类型、多媒体资源的摘要等。参考多媒体资源的任一内容信息与目标多媒体资源的匹配度用于表示该内容信息与目标多媒体资源的匹配程度,匹配度越大,表示内容信息与目标多媒体越匹配,匹配度越小,表示内容信息与目标多媒体越不匹配。由于参考多媒体资源具有多个内容信息,通过确定每个内容信息与目标多媒体资源的匹配度,以使后续能够准确地为该目标多媒体资源确定内容信息。
在一种可能实现方式中,计算机设备对任一信息进行分词处理,得到多个第二词语;根据每个第二词语在目标多媒体资源中的出现频率,获取每个第二词语与目标多媒体资源的匹配度;根据多个第二词语的权重,对多个第二词语与目标多媒体资源的匹配度进行加权,得到该任一信息与目标多媒体资源的匹 配度。
在一种可能实现方式中,如图5所示,获取内容信息与目标多媒体资源的匹配度的方式,包括以下步骤2041-2043:
2041、对任一内容信息进行分词处理,得到多个第二词语。
该步骤与上述步骤1类似,在此不再赘述。
2042、根据每个第二词语在目标多媒体资源中的出现频率,获取每个第二词语与目标多媒体资源的匹配度。
其中,第二词语与目标多媒体资源的匹配度表示第二词语与目标多媒体资源的匹配程度,第二词语在目标多媒体资源中的出现频率越大,该匹配度越大,第二词语在目标多媒体资源中的出现频率越小,该匹配度越小。
在一种可能实现方式中,该步骤2042包括:根据任一第二词语在目标多媒体资源中的第一出现频率,及该第二词语在该内容信息中的第二出现频率,获取每个第二词语与目标多媒体资源的匹配度。
在一种可能实现方式中,对于任一第二词语q i,该第二词语q i在目标资源中的第一出现频率f i、该第二词语q i在该内容信息中的第二出现频率qf i及该第二词语q i与目标多媒体资源的匹配度R(q i,d),满足以下关系:
Figure PCTCN2021075270-appb-000002
Figure PCTCN2021075270-appb-000003
其中,q i表示多个第二词语中第i个第二词语,d表示目标多媒体资源;k 1、k 2、b均为调节因子,均为任意常数,如k 1=2、b=0.75;M为调节参数,dl用于表示该目标多媒体资源的长度,如当多媒体资源为文本资源时,目标多媒体资源的长度是指文本资源包括的词语的个数,当多媒体资源为视频资源或图像资源时,目标多媒体资源的长度是指该视频资源或图像资源对应的标题及简介信息中包括的词语的个数;avg用于表示平均值,avgdl表示多个备选多媒体资源及该目标多媒体资源的平均长度。
在上述关系式中,b用于调整多媒体资源的长度对匹配度的影响,b越大,多媒体资源的长度的对第二词语与目标多媒体资源的匹配度的影响越大,b越小,多媒体资源的长度对第二词语与目标多媒体资源的匹配度的影响越小。多媒体资源的相对长度
Figure PCTCN2021075270-appb-000004
越大,则M越大,则第二词语与目标多媒体资源的匹 配度会越小。
在一种可能实现方式中,对于任一第二词语q i,该第二词语q i在该内容信息中的第二出现频率qf i为1,则该第二词语q i在目标资源中的第一出现频率f i、及该第二词语q i与目标多媒体资源d的匹配度R(q i,d),满足以下关系:
Figure PCTCN2021075270-appb-000005
2043、根据多个第二词语的权重,对多个第二词语与目标多媒体资源的匹配度进行加权处理,得到内容信息与目标多媒体资源的匹配度。
其中,第二词语的权重用于表示该第二词语与目标多媒体资源的匹配度,对内容信息与目标多媒体资源的匹配度的贡献程度,权重越大,该第二词语与目标多媒体资源的匹配度的贡献程度越大。
在一种可能实现方式中,该步骤2043包括:确定每个第二词语与目标多媒体资源的匹配度与对应的权重之间的乘积,将得到的多个乘积之间的和,确定为该内容信息与目标多媒体资源的匹配度。通过确定每个第二词语与目标多媒体资源的匹配度及每个第二词语的权重,将多个第二词语与目标多媒体资源的匹配度进行加权求和,得到该内容信息与该目标多媒体资源的匹配度。
在另一种可能实现方式中,多个第二词语与目标多媒体资源的匹配度、每个第二词语的权重及内容信息Q与目标多媒体资源d的第二匹配度Score(Q,d)满足以下关系:
Figure PCTCN2021075270-appb-000006
其中,n表示多个第二词语的总个数,n为不小于2的正整数,W i表示多个第二词语中第i个第二词语q i的权重,R(q i,d)表示第二词语q i与目标多媒体资源d的匹配度。
在一种可能实现方式中,该步骤2043包括:确定每个第二词语与目标多媒体资源的匹配度、与对应的权重之间的乘积,将得到的多个乘积之间的和与多个第二词语的个数之间的比值,确定为该内容信息与目标多媒体资源的第二匹配度。通过确定每个第二词语与目标多媒体资源的匹配度及每个第二词语的权重,将多个第二词语与目标多媒体资源的匹配度进行加权平均,得到该内容信息与该目标多媒体资源的匹配度。
对于获取第二词语的权重的方式,在一种可能实现方式中,对于任一第二 词语,根据多个多媒体资源及该第二词语,确定包含该第二词语的多媒体资源的第一数目,根据该第一数目及该多个多媒体资源的总数目,确定该第二词语的权重。其中,该多个多媒体资源包括目标多媒体资源及多个备选多媒体资源,也可以包括预设的多个多媒体资源。
在一种可能实现方式中,对于任一第二词语q i,该第一数目n(q i)、总数目N及该第二词语q i的权重W i满足以下关系:
Figure PCTCN2021075270-appb-000007
其中,c为调整参数,为任意设置的常数,如c为0.5。
在上述关系式中,在多个多媒体资源中,包含了第二词语q i的多媒体资源的第一数目数越大,则该第二词语q i在不同多媒体资源中的区分度不高,根据该第二词语q i确定与目标多媒体资源的匹配度时,该第二词语q i的重要程度越低,因此第二词语q i的权重则越低。
在一种可能实现方式中,每个第二词语q i在该内容信息中的第二出现频率qf i为1,则内容信息Q与目标多媒体资源d的匹配度Score(Q,d)满足以下关系:
Figure PCTCN2021075270-appb-000008
在另一种可能实现方式中,计算机设备根据每个信息的特征向量及目标多媒体资源的特征向量,确定每个信息与目标多媒体资源的匹配度。
在一种可能实现方式中,该步骤204包括:根据每个内容信息的特征向量及目标多媒体资源的特征向量,确定每个内容信息与目标多媒体资源的匹配度。其中,内容信息的特征向量是用于表示该内容信息的特征向量,不同的内容信息的特征向量不同。
该步骤与上述步骤202中确定每个备选多媒体资源与目标多媒体资源的匹配度的过程类似,在此不再赘述。
205、计算机设备根据多个信息中每个信息与目标多媒体资源的匹配度,从多个信息中选取目标多媒体资源的信息。
在一种可能实现方式中,在信息为内容信息的情况下,计算机设备根据多个内容信息中每个内容信息与目标多媒体资源的匹配度,从多个内容信息中选取目标多媒体资源的信息。其中,目标多媒体资源的内容信息与目标多媒体资 源的匹配度大于其他内容信息与目标多媒体资源的匹配度。
对于选取目标多媒体资源的内容信息的方式,在一种可能实现方式中,从多个内容信息中,选取与目标多媒体资源的匹配度大于第三阈值的内容信息,确定为目标多媒体资源的内容信息。
其中,第三阈值是任意设置的数值,如0.5或0.6等。内容信息与目标多媒体资源的匹配度大于第三阈值,则表示该内容信息与该目标多媒体资源相关,则将该内容信息确定为目标多媒体资源的内容信息。
在另一种可能实现方式中,将多个内容信息按照与目标多媒体资源的匹配度由小到大的顺序排列,从多个内容信息中选取匹配度最大的参考数目个内容信息,确定为参考多媒体资源。
其中,参考数目为任意设置的数值,如3或4等。由于内容信息与目标多媒体资源的匹配度越大,则表示内容信息与目标多媒体资源的匹配程度越高,因此,从多个内容信息中选取匹配度最大的参考数目个内容信息作为目标多媒体资源的内容信息。
需要说明的是,本申请实施例是以根据内容信息与目标多媒体资源的第二匹配度,确定内容信息进行说明的,而在另一实施例中,无需执行步骤203-204,可以采取其他方式,从多个内容信息中确定内容信息。
在一种可能实现方式中,计算机设备根据多个信息及每个信息的分类标签,从多个信息中选取属于目标分类标签、且与目标多媒体资源匹配的信息,将选取的信息确定为目标多媒体资源的信息。对于获取信息的分类标签的方式,在一种可能实现方式中,基于分类模型对多个信息进行处理,确定每个信息的分类标签。
对于信息为内容信息的情况,在一种可能实现方式中,根据多个内容信息及每个内容信息的分类标签,从多个内容信息中选取属于目标分类标签、且与目标多媒体资源匹配的内容信息,确定为内容信息。
其中,分类标签用于描述内容信息所属的类别,该分类标签包括劣质分类标签及优质分类标签,该劣质分类标签为低俗分类标签、谩骂分类标签和泛低质分类标签。该目标分类标签为优质分类标签,以使从多个内容信息中选取高质量的内容信息作为该内容信息。
对于获取内容信息的分类标签的方式,在一种可能实现方式中,基于分类 模型对多个内容信息进行处理,确定每个内容信息的分类标签。其中,该分类模型是训练完成的模型,用于确定内容信息的所属的类别,为内容信息生成对应的分类标签。
对于该分类模型的训练过程,获取多个样本内容信息,对该多个样本内容信息进行预处理,对预处理后的多个样本内容信息进行标注,确定每个样本内容信息的分类标签,通过预处理后的多个样本内容信息及每个样本内容信息的分类标签对该分类模型进行训练。
其中,预处理包括繁简转换、大小写转换、隐藏文字去除、低俗关键词清洗,还包括情感过滤、敏感过滤等通用过滤,以及规则判别等涉及表情、冗余字符处理与语法基础优化等,确保样本内容信息的准确性。该规则判别为过滤掉手机号、用户账号等敏感信息。
在预处理后的多个样本内容信息进行标注时,通过人工标注,确定每个样本内容信息的分类标签。该分类标签包括劣质分类标签和优质分类标签,在确定劣质分类标签时,根据内容信息的劣质程度,将劣质分类标签分为多级,不同的劣质程度对应不同的劣质分类标签;在确定优质分类标签时,根据内容信息中包括的内容及内容信息对应的点赞数目确定。
另外,该分类模型在对内容信息进行处理时,获取内容信息的特征向量,通过对内容信息的特征向量对该内容信息进行分类,从而确定内容信息的分类标签。该分类模型包括词向量获取子模型和分类子模型,该词向量获取子模型为Text CNN(Text Convolutional Neural Networks,文本卷积神经网络)模型或者其他模型,该分类子模型为SVM(Support Vector Machine,支持向量机)模型或者其他模型。
随着互联网的发展,用户在互联网中发布多媒体资源。无论是文本资源、图像资源还是视频资源,在移动互联网时代获得了飞速的发展,用户在查看多媒体资源后,还能够对多媒体资源进行评论、点赞、转发、收藏等各种互动行为。内容信息多为非正式的书面评论,存在诸多非法字符比如表情、符号等,往往需要通过信息清洗完成内容信息的规整。清洗后的内容信息基于自然语言处理相关技术(词法分析、句法分析、信息抽取、主旨话题模型)进行分析。内容信息作为用户问题、建议、态度的载体,对产品评估和改进优化极具价值。对于用户内容信息,通过文本分析解读用户的关注焦点、主要讨论话题、用户 的情感倾向,以及主要评论的主体对象等。内容信息是由此文章延伸和拓展出来的各种感悟。比如有情趣的思想火花,或者是直接拓贴,那些喜欢的段落和句子,以表示认真读过写者这些精美绝妙的文字的。不同用户的不同内容信息,相当于不同领域、不同层次、不同世界观、不同生活环境的各种用户在直接表达意见和交流。内容信息使人们讨论他们的观点并共享新信息,还能够引起人们的注意并鼓励页面浏览。
相关技术中,目标多媒体资源的内容信息是由用户发布的,或者是基于网络模型对目标多媒体资源进行处理,自动为目标多媒体资源生成内容信息,但是,这种由网络模型直接得到的内容信息的方式,得到的内容信息的准确性较差。
本申请实施例提供的方法,获取目标多媒体资源及多个备选多媒体资源的特征向量,从多个备选多媒体资源中,选取特征向量与目标多媒体资源的特征向量匹配的参考多媒体资源,以使获取到的参考多媒体资源能够与该目标多媒体资源相匹配,根据参考多媒体资源的多个信息,从多个信息中选取与目标多媒体资源匹配的信息,确定为目标多媒体资源的信息,提供了一种自动为目标多媒体资源确定信息的方式,能够保证确定信息与该目标多媒体资源相匹配,提高了信息的准确性。
通过将其他多媒体资源的信息迁移为该目标多媒体资源的信息,避免了该目标多媒体资源的信息的个数为0,优化了该目标多媒体资源的冷启动,使用户可以查看目标多媒体资源的信息,从而提高了对用户的吸引力。
通过根据多媒体资源之间的匹配度,确定与目标多媒体资源匹配的参考多媒体资源,提高了获取到的参考多媒体资源的准确性,且根据多个信息与目标多媒体资源之间的匹配度,为目标多媒体资源确定信息,提高了获取到的信息的准确性。并且,根据信息的分类标签,从多个信息中确定信息,以保证信息的质量。
图6是本申请实施例提供的一种信息管理系统的结构示意图,如图6所示,该信息管理系统包括:第一终端601、第二终端602和服务器603,第一终端601和第二终端602分别与服务器603建立通信连接。
服务器603发布多个多媒体资源供用户查看,第一终端601查看服务器603 发布的第一多媒体资源,针对第一多媒体资源生成第一内容信息,服务器603采用本申请实施例提供的方法,对第一内容信息进行迁移,将第一内容信息作为第二多媒体资源的内容信息,第二终端602在查看服务器603发布的第二多媒体资源时,查看该第二多媒体资源对应的第一内容信息。
图7是本申请实施例提供的一种信息管理系统的结构示意图,如图7所示,该信息管理系统包括:第一用户终端701、上下行内容接口服务器702、多媒体资源数据库703、调度中心服务器704、排重服务子系统705、人工审核子系统706、内容分发出口服务器707、统计上报接口服务器708、内容信息数据库709、内容信息质量评价服务器710、多媒体资源匹配服务器711、内容信息匹配服务器712、内容信息迁移服务器713、第二用户终端714及第三用户终端715。下面以信息为内容信息为例,对多媒体资源的发布过程、内容信息的发布过程以及内容信息的迁移过程进行说明。
第一阶段、多媒体资源发布阶段:
第一用户终端701为多媒体资源生成终端,该多媒体资源生成终端为PGC(Professional Generated Content,专业生产内容的机构或者组织)、UGC(User Generated Content,用户原创内容)、MCN(Multi-Channel Network,多频道网络)或者PUGC(Professional User Generated Content,专业用户原创内容)的内容生产者,通过终端或者后端接口API(Application Programming Interface,应用程序接口)系统,提供多媒体资源,如本地或者拍摄的图文内容,视频或者图集内容,在拍摄过程当中本地图文内容选择搭配的音乐,滤镜模板和图文的美化功能等,通过与上下行内容接口服务器702的通信,先获取上传服务器接口地址,然后根据该上传服务器接口地址,将多媒体资源上传至该上下行内容接口服务器702。
上下行内容接口服务器702接收到第一用户终端701上传的多媒体资源,将该多媒体资源及该多媒体资源的元信息写入多媒体资源数据库703,该多媒体资源还包括标题、发布者、摘要、封面图、发布时间,该元信息包括文件大小、封面图链接、码率、文件格式、标题、发布时间、作者、是否原创的标记等信息。并且,该上下行内容接口服务器702将上传的多媒体资源提交给调度中心服务器704。
多媒体资源数据库703用于存储多媒体资源及多媒体资源的元信息,还包括人工审核子系统706为多媒体资源确定的类别及标签信息,如,类别包括第一类别、第二类别及第三类别,第一类别是科技,第二类别是智能手机,第三类别是国内手机,标签信息包括景点XX,该多媒体资源为对景点XX进行介绍的文本资源。
调度中心服务器704负责多媒体资源流转的整个调度过程,通过上下行内容接口服务器702接收入库的多媒体资源,然后从多媒体资源数据库703中获取多媒体资源的元信息,调度人工审核子系统706和排重服务子系统705对多媒体资源进行处理,控制调度的顺序和优先级。对于多媒体资源的审核过程,先调度排重服务子系统705,对不同码率、不同清晰度、不同尺寸、部分黑屏、有无滤镜、有无Logo(标志)、在相似多媒体资源当中插入部分广告及片头片尾的裁剪都能够进行排重处理,还能够通过向量和标题是否重复的判断进行排重。排重服务子系统705用于对多媒体资源进行排重服务,在排重服务过程中,对多媒体资源进行向量化,然后建立向量的索引,然后通过比较向量之间的距离来确定相似程度;将多媒体资源通过BERT(Bidirectional Encoder Representation from Transformers,双向编码器表示模型)向量化,所有排重任务之前,先对标题短文本进行排重,对多媒体资源进行各种质量判断比如低质过滤、分类等。排重服务子系统705将排重处理的结果会写入多媒体资源数据库703,对于完全重复一样的多媒体资源不会给人工审核子系统706进行重复的二次处理。
人工审核子系统706通过读取多媒体资源数据库703中多媒体资源,通过人工来对内容是否涉及色情、赌博及政治敏感的特性进行一轮初步过滤。在初步审核的基础之上,对多媒体资源进行二次审核,是对多媒体资源进行分类和标签的标注或者确认。由于排重服务子系统705对视频资源审核的准确性不高,因此,在排重服务子系统705对视频资源审核后的基础上进行二次的人工审核处理,从而提高对视频资源的标注的准确性,且提高了处理效率。
该调度中心服务器704还将通过人工审核子系统706的多媒体资源发送至内容分发出口服务器707。
内容分发出口服务器707通过接收调度中心服务器704发送的通过人工审核子系统706的多媒体资源,将该多媒体资源以Feeds的形式进行发布,供第二用户终端714查看。
第二阶段、内容信息发布阶段:
第三用户终端715作为消费者,通过与上下行内容接口服务器702的通信,获取访问多媒体资源的索引信息,然后下载对应的多媒体资源并且通过本地播放器来播放观看。同时将上传和下载过程当中的行为数据、卡顿、加载时间、播放点击等上报给统计上报接口服务器708。并且,该第三用户终端715对多媒体资源的互动信息,如对多媒体资源的评论、点赞,转发、收藏等,上报至统计上报接口服务器708。第三用户终端715还能够对多媒体资源中的低质的内容信息进行举报,将举报的内容信息上报到内容信息数据库709,在作为样本之前经过人工审核子系统706。
统计上报接口服务器708接收第三用户终端715上传的对多媒体资源的内容信息,将该内容信息写入内容信息数据库709。
内容信息数据库709用于存储内容信息,为内容信息质量评价服务器710和内容信息迁移服务器713提供内容信息及其他互动数据。
第三阶段、内容信息迁移阶段:
内容信息质量评价服务器710按照上述实施例提供的方法,对多媒体资源的内容信息进行质量建模和分级,从而得到每个内容信息的分类标签。
多媒体资源匹配服务器711按照上述实施例提供的方法,为目标多媒体资源确定匹配的参考多媒体资源,为内容信息迁移服务器713提供匹配结果。
内容信息匹配服务器712按照上述实施例提供的方法,确定参考多媒体资源的多个内容信息中,与目标多媒体资源匹配的内容信息,为内容信息迁移服务器713提供匹配结果。
内容信息迁移服务器713通过调度服务器,根据多媒体资源匹配服务器711和内容信息匹配服务器712的匹配结果,实现内容信息迁移的效果,为目标多媒体资源生成内容信息。
该调度中心服务器704还调度内容信息迁移服务,完成优质内容信息迁移工作,同时将迁移的内容信息输出到内容分发出口服务器707,由该内容分发出口服务器707发送给第二用户终端714,供第二用户终端714查看。
图8是本申请实施例提供的一种信息确定装置的结构示意图,如图8所示,该装置包括:
特征向量获取模块801,用于获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量;
多媒体资源选取模块802,用于从多个备选多媒体资源中,选取特征向量与目标多媒体资源的特征向量匹配的参考多媒体资源;
信息确定模块803,用于根据参考多媒体资源的多个信息,从多个信息中选取与目标多媒体资源匹配的信息,将选取的信息确定为目标多媒体资源的信息。
本申请实施例提供的装置,获取目标多媒体资源及多个备选多媒体资源的特征向量,从多个备选多媒体资源中,选取特征向量与目标多媒体资源的特征向量匹配的参考多媒体资源,以使获取到的参考多媒体资源能够与该目标多媒体资源相匹配,根据参考多媒体资源的多个信息,从多个信息中选取与目标多媒体资源匹配的信息,确定为目标多媒体资源的信息,提供了一种自动为目标多媒体资源确定信息的方式,能够保证确定信息与该目标多媒体资源相匹配,提高了信息的准确性。
可选地,如图9所示,特征向量获取模块801,包括:
图网络创建单元8011,用于根据目标多媒体资源及多个备选多媒体资源,创建图网络,图网络包括目标多媒体资源对应的目标多媒体资源节点及多个备选多媒体资源对应的多个备选多媒体资源节点,满足第一关联条件的任两个多媒体资源节点连接;
第一特征向量获取单元8012,用于基于第一特征提取模型对图网络进行处理,获取目标多媒体资源节点的特征向量及多个备选多媒体资源节点的特征向量,将目标多媒体资源节点的特征向量作为目标多媒体资源的特征向量,将多个备选多媒体资源节点的特征向量作为多个备选多媒体资源的特征向量。
可选地,图网络创建单元8011,还用于对目标多媒体资源中的文本资源及多个备选多媒体资源中的文本资源进行分词处理,得到多个第一词语;根据目标多媒体资源、多个备选多媒体资源及多个第一词语,创建图网络,图网络包括目标多媒体资源对应的目标多媒体资源节点、多个备选多媒体资源对应的多个备选多媒体资源节点及多个第一词语对应的多个词节点,满足第二关联条件的词节点与多媒体资源节点连接。
可选地,如图9所示,装置还包括:
节点确定模块804,用于响应于任一第一词语在任一多媒体资源的文本资源 中的出现频率大于第一阈值,确定第一词语对应的词节点与多媒体资源对应的多媒体资源节点满足第二关联条件。
可选地,如图9所示,多媒体资源包括视频资源;特征向量获取模块801,包括:
抽帧处理单元8013,用于分别对目标多媒体资源及多个备选多媒体资源进行抽帧处理,得到目标多媒体资源对应的多个视频帧及多个备选多媒体资源对应的多个视频帧;
第二特征向量获取单元8014,用于基于第二特征提取模型,分别对目标多媒体资源对应的多个视频帧及多个备选多媒体资源对应的多个视频帧进行处理,得到目标多媒体资源的特征向量及多个备选多媒体资源的特征向量。
可选地,第二特征向量获取单元8014,还用于基于第二特征提取模型,对任一多媒体资源对应的多个视频帧分别进行处理,得到多个视频帧的特征向量;对多个视频帧的特征向量进行融合,得到多媒体资源的特征向量。
可选地,如图9所示,多媒体资源选取模块802,包括:
第一匹配度获取单元8021,用于根据目标多媒体资源的特征向量及多个备选多媒体资源的特征向量,获取每个备选多媒体资源与目标多媒体资源的匹配度;
多媒体资源选取单元8022,用于根据每个备选多媒体资源与目标多媒体资源的匹配度,从多个备选多媒体资源中选取参考多媒体资源,参考多媒体资源与目标多媒体资源的匹配度大于其他备选多媒体资源与目标多媒体资源的匹配度。
可选地,如图9所示,信息确定模块803,包括:
第二匹配度获取单元8031,用于根据参考多媒体资源的多个信息,获取多个信息中每个信息与目标多媒体资源的匹配度;
信息选取单元8032,用于根据每个信息与目标多媒体资源的匹配度,从多个信息中选取目标多媒体资源的信息,目标多媒体资源的信息与目标多媒体资源的匹配度大于其他信息与目标多媒体资源的匹配度。
可选地,第二匹配度获取单元8031,还用于对任一信息进行分词处理,得到多个第二词语;根据每个第二词语在目标多媒体资源中的出现频率,获取每个第二词语与目标多媒体资源的匹配度;根据多个第二词语的权重,对多个第 二词语与目标多媒体资源的匹配度进行加权,得到信息与目标多媒体资源的匹配度。
可选地,第二匹配度获取单元8031,还用于根据每个信息的特征向量及目标多媒体资源的特征向量,确定每个信息与目标多媒体资源的匹配度。
可选地,如图9所示,信息确定模块803,包括:
信息确定单元8033,用于根据多个信息及每个信息的分类标签,从多个信息中选取属于目标分类标签、且与目标多媒体资源匹配的信息,将选取的信息确定为目标多媒体资源的信息。
可选地,如图9所示,装置还包括:
分类标签确定模块805,用于基于分类模型对多个信息进行处理,确定每个信息的分类标签。
图10是本申请实施例提供的一种终端的结构示意图,实现上述实施例中计算机设备执行的操作。该终端1000是便携式移动终端,比如:智能手机、平板电脑、MP3播放器(Moving Picture Experts Group Audio Layer III,动态影像专家压缩标准音频层面3)、MP4(Moving Picture Experts Group Audio Layer IV,动态影像专家压缩标准音频层面4)播放器、笔记本电脑、台式电脑、头戴式设备、智能电视、智能音箱、智能遥控器、智能话筒,或其他任意智能终端。终端1000还可能被称为用户设备、便携式终端、膝上型终端、台式终端等其他名称。
通常,终端1000包括有:处理器1001和存储器1002。
处理器1001包括一个或多个处理核心,比如4核心处理器、8核心处理器等。存储器1002包括一个或多个计算机可读存储介质,该计算机可读存储介质是非暂态的,用于存储至少一个指令,该至少一个指令用于被处理器1001所具有以实现本申请中方法实施例提供的信息确定方法。
在一些实施例中,终端1000还可选包括有:外围设备接口1003和至少一个外围设备。处理器1001、存储器1002和外围设备接口1003之间通过总线或信号线相连。各个外围设备通过总线、信号线或电路板与外围设备接口1003相连。具体地,外围设备包括:射频电路1004、显示屏1005和音频电路1006中的至少一种。
射频电路1004用于接收和发射RF(Radio Frequency,射频)信号,也称电磁信号。射频电路1004通过电磁信号与通信网络及其他通信设备进行通信。
显示屏1005用于显示UI(User Interface,用户界面)。该UI包括图形、文本、图标、视频及其它们的任意组合。该显示屏1005是触摸显示屏,还用于提供虚拟按钮和/或虚拟键盘。
音频电路1006包括麦克风和扬声器。麦克风用于采集用户及环境的音频信号,并将音频信号转换为电信号输入至处理器1001进行处理,或者输入至射频电路1004以实现语音通信。出于立体声采集或降噪的目的,麦克风为多个,分别设置在终端1000的不同部位。麦克风还是阵列麦克风或全向采集型麦克风。扬声器则用于将来自处理器1001或射频电路1004的电信号转换为音频信号。
本领域技术人员理解,图10中示出的结构并不构成对终端1000的限定,包括比图示更多或更少的组件,或者组合某些组件,或者采用不同的组件布置。
图11是本申请实施例提供的一种服务器的结构示意图,该服务器1100可因配置或性能不同而产生比较大的差异,包括一个或一个以上处理器(Central Processing Units,CPU)1101和一个或一个以上的存储器1102,其中,存储器1102中存储有至少一条指令,至少一条指令由处理器1101加载并执行以实现上述各个方法实施例提供的方法。当然,该服务器还具有有线或无线网络接口、键盘及输入输出接口等部件,以便进行输入输出,该服务器还包括其他用于实现设备功能的部件,在此不做赘述。
服务器1100用于执行上述信息确定方法。
本申请实施例还提供了一种计算机设备,该计算机设备包括处理器和存储器,存储器中存储有至少一条指令,该至少一条指令由处理器加载并执行,以实现上述实施例的信息确定方法。
本申请实施例还提供了一种计算机可读存储介质,该计算机可读存储介质中存储有至少一条指令,该至少一条指令由处理器加载并执行,以实现上述实施例的信息确定方法。
本申请实施例还提供了一种计算机程序,该计算机程序中存储有至少一条指令,该至少一条指令由处理器加载并执行,以实现上述实施例的信息确定方 法。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成,也可以通过程序来指令相关的硬件完成,所述程序可以存储于一种计算机可读存储介质中,上述提到的存储介质可以是只读存储器,磁盘或光盘等。
以上所述仅为本申请实施例的可选实施例,并不用以限制本申请实施例,凡在本申请实施例的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本申请的保护范围之内。

Claims (15)

  1. 一种信息确定方法,其特征在于,所述方法包括:
    获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量;
    从所述多个备选多媒体资源中,选取特征向量与所述目标多媒体资源的特征向量匹配的参考多媒体资源;
    根据所述参考多媒体资源的多个信息,从所述多个信息中选取与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息。
  2. 根据权利要求1所述的方法,其特征在于,所述获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量,包括:
    根据所述目标多媒体资源及所述多个备选多媒体资源,创建图网络,所述图网络包括所述目标多媒体资源对应的目标多媒体资源节点及所述多个备选多媒体资源对应的多个备选多媒体资源节点,满足第一关联条件的任两个多媒体资源节点连接;
    基于第一特征提取模型对所述图网络进行处理,获取所述目标多媒体资源节点的特征向量及所述多个备选多媒体资源节点的特征向量,将所述目标多媒体资源节点的特征向量作为所述目标多媒体资源的特征向量,将所述多个备选多媒体资源节点的特征向量作为所述多个备选多媒体资源的特征向量。
  3. 根据权利要求2所述的方法,其特征在于,所述根据所述目标多媒体资源及所述多个备选多媒体资源,创建图网络,包括:
    对所述目标多媒体资源中的文本资源及所述多个备选多媒体资源中的文本资源进行分词处理,得到多个第一词语;
    根据所述目标多媒体资源、所述多个备选多媒体资源及所述多个第一词语,创建图网络,所述图网络包括所述目标多媒体资源对应的目标多媒体资源节点、所述多个备选多媒体资源对应的多个备选多媒体资源节点及所述多个第一词语对应的多个词节点,满足第二关联条件的词节点与多媒体资源节点连接。
  4. 根据权利要求3所述的方法,其特征在于,所述方法还包括:
    响应于任一第一词语在任一多媒体资源的文本资源中的出现频率大于第一阈值,确定所述第一词语对应的词节点与所述多媒体资源对应的多媒体资源节点满足所述第二关联条件。
  5. 根据权利要求1所述的方法,其特征在于,多媒体资源包括视频资源;所述获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量,包括:
    分别对所述目标多媒体资源及所述多个备选多媒体资源进行抽帧处理,得到所述目标多媒体资源对应的多个视频帧及所述多个备选多媒体资源对应的多个视频帧;
    基于第二特征提取模型,分别对所述目标多媒体资源对应的多个视频帧及所述多个备选多媒体资源对应的多个视频帧进行处理,得到所述目标多媒体资源的特征向量及所述多个备选多媒体资源的特征向量。
  6. 根据权利要求5所述的方法,其特征在于,所述基于第二特征提取模型,分别对所述目标多媒体资源对应的多个视频帧及所述多个备选多媒体资源对应的多个视频帧进行处理,得到所述目标多媒体资源的特征向量及所述多个备选多媒体资源的特征向量,包括:
    基于所述第二特征提取模型,对任一多媒体资源对应的多个视频帧分别进行处理,得到所述多个视频帧的特征向量;
    对所述多个视频帧的特征向量进行融合,得到所述多媒体资源的特征向量。
  7. 根据权利要求1所述的方法,其特征在于,所述从所述多个备选多媒体资源中,选取特征向量与所述目标多媒体资源的特征向量匹配的参考多媒体资源,包括:
    根据所述目标多媒体资源的特征向量及所述多个备选多媒体资源的特征向量,确定每个备选多媒体资源与所述目标多媒体资源的匹配度;
    根据所述每个备选多媒体资源与所述目标多媒体资源的匹配度,从所述多个备选多媒体资源中选取参考多媒体资源,所述参考多媒体资源与所述目标多媒体资源的匹配度大于其他备选多媒体资源与所述目标多媒体资源的匹配度。
  8. 根据权利要求1所述的方法,其特征在于,所述根据所述参考多媒体资源的多个信息,从所述多个信息中选取与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息,包括:
    根据所述参考多媒体资源的多个信息,获取所述多个信息中每个信息与所述目标多媒体资源的匹配度;
    根据所述每个信息与所述目标多媒体资源的匹配度,从所述多个信息中选取所述目标多媒体资源的信息,所述目标多媒体资源的信息与所述目标多媒体资源的匹配度大于其他信息与所述目标多媒体资源的匹配度。
  9. 根据权利要求8所述的方法,其特征在于,所述根据所述参考多媒体资源的多个信息,获取所述多个信息中每个信息与所述目标多媒体资源的匹配度,包括:
    对任一信息进行分词处理,得到多个第二词语;
    根据每个第二词语在所述目标多媒体资源中的出现频率,获取所述每个第二词语与所述目标多媒体资源的匹配度;
    根据所述多个第二词语的权重,对所述多个第二词语与所述目标多媒体资源的匹配度进行加权,得到所述信息与所述目标多媒体资源的匹配度。
  10. 根据权利要求8所述的方法,其特征在于,所述根据所述参考多媒体资源的多个信息,获取所述多个信息中每个信息与所述目标多媒体资源的匹配度,包括:
    根据所述每个信息的特征向量及所述目标多媒体资源的特征向量,确定所述每个信息与所述目标多媒体资源的匹配度。
  11. 根据权利要求1所述的方法,其特征在于,所述根据所述参考多媒体资源的多个信息,从所述多个信息中选取与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息,包括:
    根据所述多个信息及每个信息的分类标签,从所述多个信息中选取属于目标分类标签、且与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息。
  12. 根据权利要求11所述的方法,其特征在于,所述根据所述多个信息及每个信息的分类标签,从所述多个信息中选取属于目标分类标签、且与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息之前,所述方法还包括:
    基于分类模型对所述多个信息进行处理,确定所述每个信息的分类标签。
  13. 一种信息确定装置,其特征在于,所述装置包括:
    特征向量获取模块,用于获取目标多媒体资源的特征向量及多个备选多媒体资源的特征向量;
    多媒体资源选取模块,用于从所述多个备选多媒体资源中,选取特征向量与所述目标多媒体资源的特征向量匹配的参考多媒体资源;
    信息确定模块,用于根据所述参考多媒体资源的多个信息,从所述多个信息中选取与所述目标多媒体资源匹配的信息,将选取的信息确定为所述目标多媒体资源的信息。
  14. 一种计算机设备,其特征在于,所述计算机设备包括处理器和存储器,所述存储器中存储有至少一条指令,所述至少一条指令由所述处理器加载并执行,以实现如权利要求1至12任一权利要求所述的信息确定方法。
  15. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质中存储有至少一条指令,所述至少一条指令由处理器加载并执行,以实现如权利要求1至12任一权利要求所述的信息确定方法。
PCT/CN2021/075270 2020-03-24 2021-02-04 信息确定方法、装置、计算机设备及存储介质 WO2021190174A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/721,295 US12001474B2 (en) 2020-03-24 2022-04-14 Information determining method and apparatus, computer device, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010213262.1A CN111444357B (zh) 2020-03-24 2020-03-24 内容信息确定方法、装置、计算机设备及存储介质
CN202010213262.1 2020-03-24

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/721,295 Continuation US12001474B2 (en) 2020-03-24 2022-04-14 Information determining method and apparatus, computer device, and storage medium

Publications (1)

Publication Number Publication Date
WO2021190174A1 true WO2021190174A1 (zh) 2021-09-30

Family

ID=71654369

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/075270 WO2021190174A1 (zh) 2020-03-24 2021-02-04 信息确定方法、装置、计算机设备及存储介质

Country Status (3)

Country Link
US (1) US12001474B2 (zh)
CN (1) CN111444357B (zh)
WO (1) WO2021190174A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114417030A (zh) * 2022-01-26 2022-04-29 腾讯科技(深圳)有限公司 资源处理方法、装置、设备及计算机可读存储介质

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111444357B (zh) * 2020-03-24 2023-10-20 腾讯科技(深圳)有限公司 内容信息确定方法、装置、计算机设备及存储介质
CN111898031B (zh) * 2020-08-14 2024-04-05 腾讯科技(深圳)有限公司 一种获得用户画像的方法及装置
CN112749558B (zh) * 2020-09-03 2023-11-24 腾讯科技(深圳)有限公司 一种目标内容获取方法、装置、计算机设备和存储介质
CN112364181B (zh) * 2020-11-27 2024-05-28 深圳市慧择时代科技有限公司 一种保险产品匹配度确定方法及装置
CN112749339B (zh) * 2021-01-18 2024-05-28 陕西师范大学 一种基于旅游知识图谱的旅游路线推荐方法及系统
CN113010740B (zh) * 2021-03-09 2023-05-30 腾讯科技(深圳)有限公司 词权重的生成方法、装置、设备及介质
CN115688873A (zh) * 2021-07-23 2023-02-03 伊姆西Ip控股有限责任公司 图数据处理方法、设备及计算机程序产品
US11954436B2 (en) * 2021-07-26 2024-04-09 Freshworks Inc. Automatic extraction of situations
CN113672783B (zh) * 2021-08-11 2023-07-11 北京达佳互联信息技术有限公司 特征处理方法、模型训练方法及媒体资源处理方法
CN115600646B (zh) * 2022-10-19 2023-10-03 北京百度网讯科技有限公司 语言模型的训练方法、装置、介质及设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109358744A (zh) * 2018-08-30 2019-02-19 Oppo广东移动通信有限公司 信息共享方法、装置、存储介质及穿戴式设备
CN109359592A (zh) * 2018-10-16 2019-02-19 北京达佳互联信息技术有限公司 视频帧的处理方法、装置、电子设备及存储介质
CN110704598A (zh) * 2019-09-29 2020-01-17 北京明略软件系统有限公司 一种语句信息的抽取方法、抽取装置及可读存储介质
US20200037036A1 (en) * 2014-08-19 2020-01-30 CharacTour LLC Profiling media characters
CN111444357A (zh) * 2020-03-24 2020-07-24 腾讯科技(深圳)有限公司 内容信息确定方法、装置、计算机设备及存储介质

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180130019A1 (en) * 2016-06-21 2018-05-10 0934781 B.C. Ltd System and method for Managing user and project nodes in a graph database
US11069335B2 (en) * 2016-10-04 2021-07-20 Cerence Operating Company Speech synthesis using one or more recurrent neural networks
CN106547908B (zh) 2016-11-25 2020-03-17 三星电子(中国)研发中心 一种信息推送方法和系统
US10595039B2 (en) * 2017-03-31 2020-03-17 Nvidia Corporation System and method for content and motion controlled action video generation
US11315257B2 (en) * 2019-04-30 2022-04-26 Naidu Prakash Crj Method for real time surface tracking in unstructured environments
CN110287278B (zh) * 2019-06-20 2022-04-01 北京百度网讯科技有限公司 评论生成方法、装置、服务器及存储介质
US11093671B2 (en) * 2019-09-06 2021-08-17 Beamup Ltd. Structural design systems and methods to define areas of interest for modeling and simulation-based space planning
CN110781323A (zh) * 2019-10-25 2020-02-11 北京达佳互联信息技术有限公司 多媒体资源的标签确定方法、装置、电子设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200037036A1 (en) * 2014-08-19 2020-01-30 CharacTour LLC Profiling media characters
CN109358744A (zh) * 2018-08-30 2019-02-19 Oppo广东移动通信有限公司 信息共享方法、装置、存储介质及穿戴式设备
CN109359592A (zh) * 2018-10-16 2019-02-19 北京达佳互联信息技术有限公司 视频帧的处理方法、装置、电子设备及存储介质
CN110704598A (zh) * 2019-09-29 2020-01-17 北京明略软件系统有限公司 一种语句信息的抽取方法、抽取装置及可读存储介质
CN111444357A (zh) * 2020-03-24 2020-07-24 腾讯科技(深圳)有限公司 内容信息确定方法、装置、计算机设备及存储介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114417030A (zh) * 2022-01-26 2022-04-29 腾讯科技(深圳)有限公司 资源处理方法、装置、设备及计算机可读存储介质

Also Published As

Publication number Publication date
CN111444357B (zh) 2023-10-20
US12001474B2 (en) 2024-06-04
CN111444357A (zh) 2020-07-24
US20220237222A1 (en) 2022-07-28

Similar Documents

Publication Publication Date Title
WO2021190174A1 (zh) 信息确定方法、装置、计算机设备及存储介质
WO2022078102A1 (zh) 一种实体识别方法、装置、设备以及存储介质
US11409791B2 (en) Joint heterogeneous language-vision embeddings for video tagging and search
JP6967059B2 (ja) 映像を生成するための方法、装置、サーバ、コンピュータ可読記憶媒体およびコンピュータプログラム
CN111507097B (zh) 一种标题文本处理方法、装置、电子设备及存储介质
US20200134398A1 (en) Determining intent from multimodal content embedded in a common geometric space
CN113010703B (zh) 一种信息推荐方法、装置、电子设备和存储介质
CN112231563B (zh) 一种内容推荐方法、装置及存储介质
CN111368075A (zh) 文章质量预测方法、装置、电子设备及存储介质
CN112257661A (zh) 低俗图像的识别方法、装置、设备及计算机可读存储介质
CN111723295B (zh) 一种内容分发方法、装置和存储介质
CN114372414B (zh) 多模态模型构建方法、装置和计算机设备
CN111506794A (zh) 一种基于机器学习的谣言管理方法和装置
CN113011126B (zh) 文本处理方法、装置、电子设备及计算机可读存储介质
WO2024021685A1 (zh) 回复内容处理方法以及媒体内容互动内容的交互方法
WO2023197749A9 (zh) 背景音乐的插入时间点确定方法、装置、设备和存储介质
US20230282017A1 (en) Contextual sentiment analysis of digital memes and trends systems and methods
CN117011745A (zh) 一种数据处理方法、装置、计算机设备以及可读存储介质
CN114969282A (zh) 基于富媒体知识图谱多模态情感分析模型的智能交互方法
Chen et al. Sentiment analysis of animated film reviews using intelligent machine learning
CN116980665A (zh) 一种视频处理方法、装置、计算机设备、介质及产品
CN117009578A (zh) 视频数据的标注方法、装置、电子设备及存储介质
CN116628232A (zh) 标签确定方法、装置、设备、存储介质及产品
CN114547435B (zh) 内容质量的识别方法、装置、设备及可读存储介质
CN115168568A (zh) 一种数据内容的识别方法、装置以及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21775977

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 280223)

122 Ep: pct application non-entry in european phase

Ref document number: 21775977

Country of ref document: EP

Kind code of ref document: A1