CN112364743A - Video classification method based on semi-supervised learning and bullet screen analysis - Google Patents
Video classification method based on semi-supervised learning and bullet screen analysis Download PDFInfo
- Publication number
- CN112364743A CN112364743A CN202011204098.4A CN202011204098A CN112364743A CN 112364743 A CN112364743 A CN 112364743A CN 202011204098 A CN202011204098 A CN 202011204098A CN 112364743 A CN112364743 A CN 112364743A
- Authority
- CN
- China
- Prior art keywords
- bullet screen
- model
- video
- classification
- dataset
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000004458 analytical method Methods 0.000 title claims description 7
- 238000012549 training Methods 0.000 claims abstract description 35
- 230000008451 emotion Effects 0.000 claims abstract description 15
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 6
- 230000003993 interaction Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 10
- 238000012545 processing Methods 0.000 description 7
- 238000013145 classification model Methods 0.000 description 6
- 238000004590 computer program Methods 0.000 description 6
- 230000002996 emotional effect Effects 0.000 description 6
- 238000002372 labelling Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 238000003058 natural language processing Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/75—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/783—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/7844—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using original textual content or text extracted from visual content or transcript of audio data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/12—Use of codes for handling textual entities
- G06F40/14—Tree-structured documents
- G06F40/143—Markup, e.g. Standard Generalized Markup Language [SGML] or Document Type Definition [DTD]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/47—End-user applications
- H04N21/478—Supplemental services, e.g. displaying phone caller identification, shopping application
- H04N21/4788—Supplemental services, e.g. displaying phone caller identification, shopping application communicating with other users, e.g. chatting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Library & Information Science (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Signal Processing (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The barrage video is a product of entertainment interaction emerging in recent years, the prior knowledge can be used for classification of the barrage video, emotion information in the barrage video is mined, and prediction is made on video content. The method comprises the following steps: firstly, acquiring and preprocessing a bullet screen data set. Secondly, a semi-supervised learning mode is used for extracting emotion and theme information in the bullet screen by using a small part of labeled data and a large amount of unlabeled data training models. And thirdly, detecting the content of the video according to the result, using the result in combination with a time axis to generate a linear sequence containing emotion labels and theme labels, and completing a video classification task by comparing the sequence similarity of different videos.
Description
Technical Field
The application belongs to the technical field of video processing, and particularly relates to a video classification method and device of a bullet screen data processing method based on a deep learning model BERT and semi-supervised learning.
Background
In recent years, watching online videos has become a main way for mass entertainment, and the communication way of watching video bulletin screens is more and more popular with people. People participate in this interactive mode while generating huge amounts of barrage data and video data. However, how to mine valuable information in massive bullet screen data is a problem faced at present.
A video classification task belongs to the technical field of video processing, and the traditional video classification task is mainly completed by using a machine learning method and comprises the following steps: training of classification models using annotation data. Specifically, the method comprises the following steps: extracting a key frame sequence from a video, carrying out multi-dimensional embedding processing on image content characteristics to obtain a multi-dimensional video content characteristic vector of the target video, sending the characteristic vector into a neural network model, and learning characteristics of a sample from a large amount of training data. And predicting a final classification of the actual data using the trained model.
Barrage video is a product of entertainment interaction emerging in recent years, and refers to a barrage containing a time sequence feature embedded in video, and the barrage often contains understanding of video content by a video viewer (user). For the classification of the barrage video, the barrage information can be used, the emotion information in the barrage information is mined, the prediction is made on the video content, and the classification problem of the video is solved from the viewpoint of barrage analysis by applying the theory and the method of Natural Language Processing (NLP).
Disclosure of Invention
In order to solve the technical problems, the application provides a video classification method based on semi-supervised learning and barrage analysis, which can extract emotion information in a barrage so as to detect the content of a video, and uses the result in combination with a time axis to generate a linear sequence containing emotion labels and theme labels, and completes a video classification task by comparing the sequence similarity of different videos. And finally obtaining a classification result. The method has better accuracy on the classification result of the bullet screen video, is a supplement to the traditional video classification method, and has certain practical value.
According to an aspect of the present application, a video classification method based on semi-supervised learning and barrage analysis is provided, the method including:
and S1, acquiring bullet screen data and preprocessing the bullet screen to construct a bullet screen data set. And performing word segmentation on the processed bullet screen data of the training set by using jieba word segmentation, marking sentences by using a dictionary encoder, generating an input sequence and vectorizing and expressing the input sequence.
And S2, sending the spliced feature vectors into a pre-training language model BERT, and updating specific parameters in a subjective and objective bullet screen classification task, an emotion multi-classification task and a theme classification task in a deep space through learning of a marked bullet screen data set L. The Teacher model was obtained.
And S3, labeling the unmarked data set by using a Teacher model to generate a pseudo label data set P.
S4, training a larger model Student on L + P, and adding noise data to the dataset before training, this process may force the new model to be insensitive to noise data.
And S5, returning to S4, taking the Student model as a Teacher model, continuously labeling the unmarked data set to generate a new pseudo label data set P, and obtaining a new Student model until the model converges or the computing resources are exhausted.
And S6, classifying the test samples by using the trained model, fusing the time characteristics with the classification result to obtain corresponding graph structures of different videos, obtaining corresponding sequences through traversal, and finishing video classification by comparing the sequence similarity of the different videos.
Specifically, step S1 specifically includes:
s101, special characters such as expressions are transferred into characters. And the data set was divided into a training set and a test set at a ratio of 9 to 1.
S102, marking the bullet screen data of the test set, wherein the marking is divided into two parts, including: and classifying and labeling the subjective and objective bullet screen data, and dividing the bullet screen into a subjective bullet screen and a bullet screen without the subjective emotional information and an objective bullet screen according to whether the bullet screen contains the subjective emotional information.
S103, subjective barrage emotion multi-classification marking, such as: label like happy, anger, grief, happy, frightened, thinking, fear, etc. And custom bullet screen theme annotations, such as: cook, fight, chat, kissing, etc. Approximately annotate the bullet screen data of the training set 1/10.
S104, one word or character per action in the dictionary D.
S105, marking mode is as follows: connecting a [ CLS ] mark at the beginning of a sentence, and adding an [ SEP ] mark between an auxiliary sentence and an original sentence, wherein the [ SEP ] mark specifically comprises the following steps: [ CLS ] original sentence sequence [ SEP ] auxiliary sentence sequence [ SEP ].
S106, vectorization representation specifically comprises the following steps: word vectors, position vectors, and segment vectors.
Specifically, step S2 specifically includes:
s201, the Teacher model uses a BERT pre-training language model published by Google. The parameters used were: the transform _ block is 6, the Embedding _ distribution is 384, the num _ headers is 12, and the TotalParameters is 23M. The pre-training anticipates used the Chinese Wikipedia anticipates nlp _ chinese _ corpus.
S202, adding dynamic padding to the model to optimize training speed.
Specifically, step S4 specifically includes:
s401, the same Student model uses the BERT pre-training language model published by Google. But here the Student model is larger than the Teacher model in order to maximize the training. The parameters used were: 12 for transform _ block, 768 for embedded _ dimension, 12 for num _ headers, 110M for TotalParameters
S402, adding noise data to the data set L + P before training the Student model is a data enhancement transformation. The specific implementation in the bullet screen text is back _ translation and TF-IDF word replacement.
Specifically, step S5 specifically includes:
s501, reducing the consistency loss to the minimum gradually propagates the tag information from the tagged data to the untagged data.
Specifically, step S6 specifically includes:
s601, dividing bullet screen data into two types according to whether subjective emotional information is contained or not by using a subjective and objective classification model. And further, the subjective bullet screen is sent into a multi-emotion classification model for prediction, and the subjective prediction is carried out on the objective bullet screen.
S602, in actual prediction, because the user' S view and feel of the same video at the same time and the emotion expressed by the launched barrage are not completely consistent, multiple emotions or topics may be predicted at the same time of the same time axis, here, a directed graph is used to record classification results, and votes of the prediction results are used as weights of directed graph edges. The predicted result is taken as a point. Finally, a bullet screen emotion-time digraph and a bullet screen theme-time digraph are obtained.
And S603, traversing the directed graph, and selecting four modes of breadth-first search (BFS), depth-first search (DFS), breadth-first search (WBFS) based on weight and depth-first search (WDFS) based on weight for traversing. Tests have shown that weight-based depth first search (WDFS) is better in the actual classification effect.
S604, for the two sequences, the similarity can be determined by comparing the Levenshtein (Levenshtein) distance of the two sequences of different videos. And finally finishing the classification task of the video.
The invention has the beneficial effects that: mainly comprises the following aspects:
firstly, the method uses the theoretical knowledge of NLP to complete the video classification task in the technical field of video processing, and compared with the traditional video classification method, the method is an attempt to completely classify the video with the barrage from another angle. And certain effect is obtained in practical application.
Secondly, when the BERT model used by the method is proposed by Google, the BERT model is pre-trained on a large number of text data sets, and compared with models such as CNN, RNN and LSTM, the method can reduce pre-training steps and reduce complex workload.
And thirdly, a semi-supervised learning mode is used, so that the defect of small labeled data amount is overcome, and more features can be learned from massive unlabeled data sets. On the other hand, the overfitting phenomenon which is easy to occur under the condition that the data quantity is insufficient in supervised learning is also solved.
And fourthly, the video classification model uses the prior knowledge of the user, so that the effect in the video classification task with the bullet screen is better.
The above and other objects, advantages and features of the present application will become more apparent to those skilled in the art from the following detailed description of specific embodiments thereof, taken in conjunction with the accompanying drawings.
Drawings
Some specific embodiments of the present application will be described in detail hereinafter by way of illustration and not limitation with reference to the accompanying drawings. The same reference numbers in the drawings identify the same or similar elements or components. Those skilled in the art will appreciate that the drawings are not necessarily drawn to scale. In the drawings:
FIG. 1 is a schematic flow chart diagram of a video classification method based on semi-supervised learning and barrage analysis according to one embodiment of the present application;
FIG. 2 is a schematic flow chart diagram of model training according to one embodiment of the present application;
FIG. 3 is a diagram of a model architecture according to one embodiment of the present application;
Detailed Description
The implementation process mainly comprises two steps: training a network by using the preprocessed bullet screen data; and then classifying the actual bullet screen video by using the trained model.
And S1, acquiring bullet screen data and preprocessing the bullet screen to construct a bullet screen data set. And performing word segmentation on the processed bullet screen data of the training set by using jieba word segmentation, marking sentences by using a dictionary encoder, generating an input sequence and vectorizing and expressing the input sequence.
And S2, sending the spliced feature vectors into a pre-training language model BERT, and updating specific parameters in a subjective and objective bullet screen classification task, an emotion multi-classification task and a theme classification task in a deep space through learning of a marked bullet screen data set L. The Teacher model was obtained.
And S3, labeling the unmarked data set by using a Teacher model to generate a pseudo label data set P.
S4, training a larger model Student on L + P, and adding noise data to the dataset before training, this process may force the new model to be insensitive to noise data.
And S5, returning to S4, taking the Student model as a Teacher model, continuously labeling the unmarked data set to generate a new pseudo label data set P, and obtaining a new Student model until the model converges or the computing resources are exhausted.
And S6, classifying the test samples by using the trained model, fusing the time characteristics with the classification result to obtain corresponding graph structures of different videos, obtaining corresponding sequences through traversal, and finishing video classification by comparing the sequence similarity of the different videos.
Specifically, step S1 specifically includes:
s101, special characters such as expressions are transferred into characters. And the data set was divided into a training set and a test set at a ratio of 9 to 1.
S102, marking the bullet screen data of the test set, wherein the marking is divided into two parts, including: and classifying and labeling the subjective and objective bullet screen data, and dividing the bullet screen into a subjective bullet screen and a bullet screen without the subjective emotional information and an objective bullet screen according to whether the bullet screen contains the subjective emotional information.
S103, subjective barrage emotion multi-classification marking, such as: label like happy, anger, grief, happy, frightened, thinking, fear, etc. And custom bullet screen theme annotations, such as: cook, fight, chat, kissing, etc. Approximately annotate the bullet screen data of the training set 1/10.
S104, one word or character per action in the dictionary D.
S105, marking mode is as follows: connecting a [ CLS ] mark at the beginning of a sentence, and adding an [ SEP ] mark between an auxiliary sentence and an original sentence, wherein the [ SEP ] mark specifically comprises the following steps: [ CLS ] original sentence sequence [ SEP ] auxiliary sentence sequence [ SEP ].
S106, vectorization representation specifically comprises the following steps: word vectors, position vectors, and segment vectors.
Specifically, step S2 specifically includes:
s201, the Teacher model uses a BERT pre-training language model published by Google. The parameters used were: the transform _ block is 6, the Embedding _ distribution is 384, the num _ headers is 12, and the TotalParameters is 23M. The pre-training anticipates used the Chinese Wikipedia anticipates nlp _ chinese _ corpus.
S202, adding dynamic padding to the model to optimize training speed.
Specifically, step S4 specifically includes:
s401, the same Student model uses the BERT pre-training language model published by Google. But here the Student model is larger than the Teacher model in order to maximize the training. The parameters used were: 12 for transform _ block, 768 for embedded _ dimension, 12 for num _ headers, 110M for TotalParameters
S402, adding noise data to the data set L + P before training the Student model is a data enhancement transformation. The specific implementation in the bullet screen text is back _ translation and TF-IDF word replacement.
And S403, adopting a consistency training mode in the whole training process. Firstly, enhancing the non-label data, then sending the data before enhancement and the data after enhancement into the network, obtaining a prediction result, calculating the two results into a KL divergence as unsupervised cross entropy loss, and adding the unsupervised consistency loss together for back propagation.
Further, the objective function used in step S403 is as follows:
the former term is the loss with label, the latter term is the loss without label, and λ represents the ratio between the two.
Specifically, step S6 specifically includes:
s601, dividing bullet screen data into two types according to whether subjective emotional information is contained or not by using a subjective and objective classification model. And further, the subjective bullet screen is sent into a multi-emotion classification model for prediction, and the subjective prediction is carried out on the objective bullet screen.
S602, in actual prediction, because the user' S view and feel of the same video at the same time and the emotion expressed by the launched barrage are not completely consistent, multiple emotions or topics may be predicted at the same time of the same time axis, here, a directed graph is used to record classification results, and votes of the prediction results are used as weights of directed graph edges. The predicted result is taken as a point. Finally, a bullet screen emotion-time digraph and a bullet screen theme-time digraph are obtained.
And S603, traversing the directed graph, and selecting four modes of breadth-first search (BFS), depth-first search (DFS), breadth-first search (WBFS) based on weight and depth-first search (WDFS) based on weight for traversing. Tests have shown that weight-based depth first search (WDFS) is better in the actual classification effect.
And S604, determining the similarity of different videos by comparing the Levenshtein (Levenshtein) distances of the two sequences of the different videos. And finishing the classification task of the videos according to the similarity scores of different videos.
Further, the calculation formula of the levenstan distance of the two sequences in step S604 is as follows:
where a, b are two strings and i/j is the array index.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (6)
1. A video classification method based on semi-supervised learning and barrage analysis comprises the following steps:
and acquiring bullet screen data, preprocessing the bullet screen and constructing a bullet screen data set. And performing word segmentation on the processed bullet screen data of the training set by using jieba word segmentation, marking sentences by using a dictionary encoder, generating an input sequence and vectorizing and expressing the input sequence.
2. The method according to claim 1, wherein the sequence feature vector is fed into a pre-training language model BERT, and specific parameters in the subjective and objective bullet screen classification task, the emotion multi-classification task and the theme classification task in the deep space are updated through learning of the labeled bullet screen data set L. The Teacher model was obtained.
3. The method of claim 2, wherein the Teacher model labels the unlabeled dataset to produce a pseudo-labeled dataset P.
4. The method of claim 3, wherein a larger model Student is trained on the L + P dataset and noise data is added to the dataset before training.
5. The method of claims 3 to 4, wherein the Student model continues to label the unlabeled dataset as the Teacher model to generate a new pseudo label dataset P and obtain a new Student model until the model converges or the computing resources are exhausted.
6. The method as claimed in claims 1 to 5, wherein the trained classifier classifies the test samples, integrates the time characteristics with the classification results to obtain the corresponding graph structures of different videos, obtains the corresponding sequences through traversal, and completes video classification by comparing the sequence similarity of different videos.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011204098.4A CN112364743A (en) | 2020-11-02 | 2020-11-02 | Video classification method based on semi-supervised learning and bullet screen analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011204098.4A CN112364743A (en) | 2020-11-02 | 2020-11-02 | Video classification method based on semi-supervised learning and bullet screen analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112364743A true CN112364743A (en) | 2021-02-12 |
Family
ID=74513347
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011204098.4A Pending CN112364743A (en) | 2020-11-02 | 2020-11-02 | Video classification method based on semi-supervised learning and bullet screen analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112364743A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113177138A (en) * | 2021-04-30 | 2021-07-27 | 南开大学 | Supervised video classification method based on bullet screen and title analysis |
| CN114880478A (en) * | 2022-06-07 | 2022-08-09 | 昆明理工大学 | Weak supervision aspect category detection method based on theme information enhancement |
| CN116128768A (en) * | 2023-04-17 | 2023-05-16 | 中国石油大学(华东) | Unsupervised image low-illumination enhancement method with denoising module |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018014610A1 (en) * | 2016-07-20 | 2018-01-25 | 武汉斗鱼网络科技有限公司 | C4.5 decision tree algorithm-based specific user mining system and method therefor |
| CN110399490A (en) * | 2019-07-17 | 2019-11-01 | 武汉斗鱼网络科技有限公司 | A kind of barrage file classification method, device, equipment and storage medium |
| CN110569354A (en) * | 2019-07-22 | 2019-12-13 | 中国农业大学 | Barrage emotion analysis method and device |
| CN111860237A (en) * | 2020-07-07 | 2020-10-30 | 中国科学技术大学 | Video emotion fragment identification method and device |
-
2020
- 2020-11-02 CN CN202011204098.4A patent/CN112364743A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018014610A1 (en) * | 2016-07-20 | 2018-01-25 | 武汉斗鱼网络科技有限公司 | C4.5 decision tree algorithm-based specific user mining system and method therefor |
| CN110399490A (en) * | 2019-07-17 | 2019-11-01 | 武汉斗鱼网络科技有限公司 | A kind of barrage file classification method, device, equipment and storage medium |
| CN110569354A (en) * | 2019-07-22 | 2019-12-13 | 中国农业大学 | Barrage emotion analysis method and device |
| CN111860237A (en) * | 2020-07-07 | 2020-10-30 | 中国科学技术大学 | Video emotion fragment identification method and device |
Non-Patent Citations (1)
| Title |
|---|
| 洪庆;王思尧;赵钦佩;李江峰;饶卫雄;: "基于弹幕情感分析和聚类算法的视频用户群体分类", 计算机工程与科学, no. 06, 15 June 2018 (2018-06-15) * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113177138A (en) * | 2021-04-30 | 2021-07-27 | 南开大学 | Supervised video classification method based on bullet screen and title analysis |
| CN114880478A (en) * | 2022-06-07 | 2022-08-09 | 昆明理工大学 | Weak supervision aspect category detection method based on theme information enhancement |
| CN114880478B (en) * | 2022-06-07 | 2024-04-23 | 昆明理工大学 | Weak supervision aspect category detection method based on theme information enhancement |
| CN116128768A (en) * | 2023-04-17 | 2023-05-16 | 中国石油大学(华东) | Unsupervised image low-illumination enhancement method with denoising module |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113254599B (en) | Multi-label microblog text classification method based on semi-supervised learning | |
| Qiu et al. | DGeoSegmenter: A dictionary-based Chinese word segmenter for the geoscience domain | |
| US20220245365A1 (en) | Translation method and apparatus based on multimodal machine learning, device, and storage medium | |
| Chang et al. | Research on detection methods based on Doc2vec abnormal comments | |
| CN112364743A (en) | Video classification method based on semi-supervised learning and bullet screen analysis | |
| CN111061861B (en) | Text abstract automatic generation method based on XLNet | |
| CN108108468A (en) | A kind of short text sentiment analysis method and apparatus based on concept and text emotion | |
| CN110263165A (en) | A kind of user comment sentiment analysis method based on semi-supervised learning | |
| CN113392209A (en) | Text clustering method based on artificial intelligence, related equipment and storage medium | |
| CN114595327A (en) | Data enhancement method and device, electronic equipment and storage medium | |
| CN112131430A (en) | Video clustering method and device, storage medium and electronic equipment | |
| CN111897954A (en) | User comment aspect mining system, method and storage medium | |
| CN111382231A (en) | Intention recognition system and method | |
| CN113392179A (en) | Text labeling method and device, electronic equipment and storage medium | |
| CN114722805A (en) | Little sample emotion classification method based on size instructor knowledge distillation | |
| He et al. | Deep learning in natural language generation from images | |
| Jishan et al. | Natural language description of images using hybrid recurrent neural network | |
| Nassiri et al. | Arabic L2 readability assessment: Dimensionality reduction study | |
| CN115062174A (en) | End-to-end image subtitle generating method based on semantic prototype tree | |
| CN110765241A (en) | Super-outline detection method and device for recommendation questions, electronic equipment and storage medium | |
| Tüselmann et al. | Recognition-free question answering on handwritten document collections | |
| CN113486143A (en) | User portrait generation method based on multi-level text representation and model fusion | |
| Tashu et al. | Deep learning architecture for automatic essay scoring | |
| CN112749566A (en) | English writing auxiliary oriented semantic matching method and device | |
| CN114997175A (en) | Emotion analysis method based on field confrontation training |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |