A video recommendation method based on content extraction and rating prediction

Abstract

The present invention relates to a video recommendation method based on content extraction and rating prediction. This method performs text analysis on video content descriptions, extracts information such as video keywords, converts video features into text vectors, and calculates a content similarity matrix between different videos. Furthermore, relevant keyword information is used to construct a knowledge graph, and path-based knowledge reasoning is employed to calculate the correlation between adjacent and non-adjacent video content entities. A review-aware algorithm is used to predict user ratings for videos, using the predicted ratings as an enhancement factor. The similarity matrix, derived from the fusion of content extraction and knowledge reasoning, is then calculated to produce a recommendation model.

Description

Technical Field

[0001] The present invention relates to the field of data mining, and mainly relates to video recommendation based on content extraction and rating prediction. Background Art

[0002] With the widespread dissemination of the Internet and the rapid growth of online information, video programs have become the main type of media resources that people receive in their daily lives. However, due to the explosive production and promotion of new video resources, the public is already in an environment of information overload. It is difficult to effectively discover satisfactory high-quality video resources by relying solely on their own experience to actively use classification or search to query.

[0003] For IPTV-related film and television content products, users can currently browse or order them on both mobile and TV devices. However, the display of product resources is mainly configured by operators and relies on their individual experience. The scope or direction of product selection is relatively fixed and narrow, lacking flexibility, and unable to push personalized content to users to further improve user stickiness and conversion rate.

[0004] As an effective means of information filtering and content push, the recommendation system can enable users to passively obtain content of potential interest through a mutual recommendation mechanism between contents.

[0005] Common recommendation methods primarily analyze and recommend videos based on historical user behavior data, using machine learning models or matrix decomposition to identify videos that might interest users. However, these methods often fail to provide easy-to-understand explanations (i.e., reasons for the recommendation) during the recommendation process. Furthermore, algorithms that use comments as recommendation information often use text analysis to extract video-related content tags as user preference factors for recommendation calculations across different videos, while ignoring the emotional expressions of users during the evaluation process.

[0006] Therefore, a solution is needed to recommend videos that users are interested in and have better quality evaluation. Summary of the Invention

[0007] This Summary is provided to introduce some concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

[0008] This method performs text analysis on video content descriptions, extracting information such as video keywords, converting video features into text vectors, and calculating a content similarity matrix between different videos. It also constructs a knowledge graph using relevant keyword information and employs path-based knowledge reasoning to calculate correlations between adjacent and non-adjacent video content entities. It also uses a review-aware algorithm to predict user ratings for videos, using the predicted ratings as a boosting factor. The similarity matrix derived from the fusion of content extraction and knowledge reasoning is then calculated to create a recommendation model.

[0009] According to one embodiment of the present invention, a video recommendation method based on content extraction and rating prediction is disclosed, including: extracting text content of a video to generate a keyword set of the video content; constructing a content extraction-based similarity matrix based on the keyword set of the video content through a text similarity algorithm; constructing a knowledge graph based on the keyword set of the video content, and performing knowledge reasoning on the knowledge graph through a path-based translation model to construct a knowledge reasoning-based similarity matrix; analyzing video comments with a comment-aware type algorithm, and extracting multidimensional features including content attributes and emotional attributes, and predicting the video to obtain a prediction score; and fusing the constructed content extraction-based similarity matrix and the constructed knowledge reasoning-based similarity matrix through a fusion factor to obtain a final recommendation model, wherein the fusion factor is based on the prediction score.

[0010] According to another embodiment of the present invention, a video recommendation system based on content extraction and rating prediction is disclosed, comprising: a video content similarity matrix acquisition module, comprising a similarity matrix acquisition module based on content extraction and a similarity matrix acquisition module based on knowledge reasoning, wherein the video content similarity matrix acquisition module is configured to extract the text content of the video to generate a keyword set of the video content, wherein the similarity matrix acquisition module based on content extraction is configured to construct a similarity matrix based on content extraction by a text similarity algorithm based on the keyword set of the video content, and wherein the similarity matrix acquisition module based on knowledge reasoning is configured to extract the similarity matrix based on the keyword set of the video content A keyword set is used to construct a knowledge graph, and knowledge reasoning is performed on the knowledge graph through a path-based translation model to construct a similarity matrix based on knowledge reasoning; a prediction score acquisition module is configured to analyze the comments of the video using a comment-aware type algorithm, and at the same time extract multidimensional features including content attributes and emotional attributes, and predict the video to obtain a prediction score; and a recommendation model generation module is configured to fuse the constructed similarity matrix based on content extraction and the constructed similarity matrix based on knowledge reasoning through a fusion factor to obtain a final recommendation model, wherein the fusion factor is based on the prediction score.

[0011] According to another embodiment of the present invention, a computing device for video recommendation based on content extraction and rating prediction is disclosed, including: a processor; a memory, wherein the memory stores instructions, and the instructions can perform the method described above when executed by the processor.

[0012] These and other features and advantages will become apparent from reading the following detailed description and referring to the associated drawings.It is to be understood that both the foregoing general description and the following detailed description are illustrative only and are not restrictive of the aspects claimed. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order that the manner in which the above-mentioned features of the present invention are understood in detail, a more particular description of the contents briefly summarized above may be given with reference to various embodiments, some aspects of which are illustrated in the accompanying drawings. It should be noted, however, that the accompanying drawings illustrate only certain typical aspects of the invention and are not to be considered limiting of its scope, as the description may admit to other equally effective aspects.

[0014] Figure 1 FIG2 shows a block diagram of a video recommendation system 100 based on content extraction and rating prediction according to an embodiment of the present invention;

[0015] Figure 2 A flowchart of a method 200 for video recommendation based on content extraction and rating prediction according to an embodiment of the present invention is shown; and

[0016] Figure 3 1 shows a block diagram 300 of an exemplary computing device according to one embodiment of the present invention. DETAILED DESCRIPTION

[0017] The present invention will be described in detail below with reference to the accompanying drawings, and the features of the present invention will be further apparent in the following specific description.

[0018] The following is an explanation of the terms used in the present invention, which include general meanings well known to those skilled in the art:

[0019] Knowledge graph: A graph-based data structure in which each node represents an entity (which can be a specific person, event, object, or abstract concept), and each edge represents a relationship between entities. A triple consisting of {entity, relationship, entity} is the basic unit of representation or storage in a knowledge graph.

[0020] TextRank algorithm: A graph-based ranking algorithm for keyword extraction and document summarization. It uses the co-occurrence information (semantics) between words within a document to extract keywords. It can extract keywords and keyword phrases from a given text and use extractive automatic summarization methods to extract key sentences from the text.

[0021] The PTransE algorithm is an extension of the TransE algorithm. Based on a distributed vector representation of entities and relationships, the TransE algorithm considers triple relationships as a translation from the head entity to the tail entity. The ultimate goal is to represent all entities and relationships in the knowledge base as low-dimensional vectors. However, the TransE algorithm only considers direct relationships between entities. In reality, much information cannot be derived through direct reasoning, so single-step reasoning is not applicable. The PTransE algorithm instead uses a multi-step reasoning approach for knowledge reasoning.

[0022] SentiRec algorithm: A review-aware recommendation model that incorporates review sentiment when modeling users and products. The SentiRec approach consists of two steps: 1. Incorporating review sentiment, which involves encoding each review into a fixed-length review vector that reflects the review's sentiment; 2. Training the review vector to reflect the sentiment of the review text and generate recommendations.

[0023] CNN: Convolutional Neural Networks (CNN) is a type of feedforward neural network that includes convolution calculations and has a deep structure. It is one of the representative algorithms of deep learning.

[0024] Currently, IPTV video content is primarily configured and displayed by operators through backend configuration. Access to content is limited to user-defined search and global search, making it difficult to proactively provide personalized video recommendations. Recommendations often lack explicit reasons for the recommendations (i.e., explainability), resulting in a poor user experience. The range of recommended videos is limited, and product recommendations often lack coverage. Furthermore, the emotional expressions in user reviews are often overlooked in the recommendation process. Therefore, incorporating actual user feedback into the recommendation process, with video content analysis as a primary factor, could improve recommendation effectiveness.

[0025] The present invention combines the content information of the video itself with the user's comment information, and adopts technical methods such as text data analysis, knowledge graph path reasoning and comment perception algorithm model to improve the interpretability of recommendations through knowledge reasoning, and enhances it with the predicted scores of user comments with implicit emotional expressions, recommending results with similar content and good reputation to users, thereby improving the effectiveness of recommendations.

[0026] Figure 1 FIG. 1 shows a block diagram of a video recommendation system 100 based on content extraction and rating prediction according to an embodiment of the present invention. Figure 1 As shown in FIG, the system 100 is divided into modules, and the modules communicate and exchange data with each other in a manner known in the art. In the present invention, each module can be implemented by software or hardware or a combination thereof. Figure 1 As shown, the system 100 includes a video content similarity matrix acquisition module 101, a prediction score acquisition module 102, and a recommendation model generation module 103. The video content similarity matrix acquisition module 101 further includes a content extraction-based similarity matrix acquisition module 104 and a knowledge reasoning-based similarity matrix acquisition module 105. The system 100 can be implemented on a server (e.g., a computing device) for providing video recommendations, such as a backend server of an IPTV service, or can be implemented in a distributed manner on a cloud for providing video recommendations, or can be implemented in a local computing device of a user device.

[0027] In general, the video content similarity matrix acquisition module 101 is configured to construct two matrices: a content-extraction-based similarity matrix and a knowledge-inference-based similarity matrix. The prediction score acquisition module 102 is configured to calculate a predicted video score using the SentiRec algorithm, which is based on a convolutional neural network. The recommendation model generation module 103 is configured to fuse the content-extraction-based similarity matrix and the knowledge-inference-based similarity matrix constructed by the video content similarity matrix acquisition module 101, and enhance them using a comment perception factor derived from the predicted video score calculated by the prediction score acquisition module 102, thereby generating a recommendation model.

[0028] According to one embodiment of the present invention, the video content similarity matrix acquisition module 101 is configured to extract the text content of the target video to generate a keyword set for the video content. According to one embodiment of the present invention, the scope of the target video is not specifically limited and can refer to all videos in the business system or one or more types of videos filtered according to specific business requirements.

[0029] Specifically, for target videos that lack text descriptions, we first extract key frames and then use OCR (Optical Character Recognition) tools to identify the text subtitle information corresponding to the target video. Next, we use the Jieba word segmentation tool to segment the description information of each video (including text information such as the theme, director, actors, and introduction), removing invalid words such as stop words. Then, we use the TF-IDF algorithm, i.e., the calculation method in the following formula (1), to calculate the weights of the segmented corpus and extract keywords to form a keyword set for the video content.

[0030]

[0031] Among them TF ij is the word frequency probability after word segmentation. For example, in the description of a movie, the occurrence rate of "science fiction" is 0.16, and IDF i It is the inverse document index of the corresponding keyword in the collection.

[0032] Although the above description shows that the TF-IDF algorithm is used to extract keywords, those skilled in the art will appreciate that the TextRank algorithm, for example, may also be used to extract keywords.

[0033] According to one embodiment of the present invention, the content extraction-based similarity matrix acquisition module 104 is configured to construct a content extraction-based similarity matrix based on a keyword set of the video content using a text similarity algorithm. Specifically, the word frequency vector V of each video text content is calculated using the keyword set. i =(P 1,i ,P 2,i ,P 3,i ,P 4,i ,P 5,i …,P k,i ), the cosine similarity is calculated using formula (2), and the similarity of different video text vectors (i.e., video similarity) is calculated to form a similarity matrix based on content extraction.

[0034]

[0035] According to one embodiment of the present invention, the knowledge reasoning-based similarity matrix acquisition module 105 is configured to construct a knowledge graph based on a keyword set of the video content, and perform knowledge reasoning through a path-based translation model to expand entity relationships and generate a knowledge reasoning-based similarity matrix.

[0036] Specifically, the knowledge graph is constructed based on the keyword set extracted from the video content, and the knowledge graph ontology construction and relationship filling are supplemented and improved by an external crawler program. Generally speaking, the target entity and its related attributes can be obtained through the keyword set extracted from the video content. After that, different entities are associated to obtain a triple for constructing the knowledge graph, such as {entity, relationship, entity}. In the context of the present invention, the entity can be the video name, and the relationship can be the association relationship between the video and other videos.

[0037] In practice, due to the potential limitations of the target video in terms of type, content, etc., the keyword set extracted from the target video may not be sufficient to construct a complete knowledge graph. Therefore, a crawler program can be used to further extract media resource keywords from external data sources (for example, data sources including structured information or semi-structured information, such as Baidu Encyclopedia, Douban Film and Television, and other media resource information data sources) to construct more substantial triples for constructing knowledge graphs.

[0038] Next, using the PTransE data model, we applied the PRA path ranking algorithm to the relational entities in the knowledge graph, discovering paths between non-adjacent entities. During model training, to reduce duplicate data, we set the length of the random walk to l, and added new triples with a score of at least 0.05 to the entity set. We calculated the similarity between video content based on knowledge reasoning using formula (3).

[0039]

[0040] Here, S(v1,p,v2) represents the similarity measure between two entities under semantic meaning p, calculated using the Euclidean distance. P(p|v1,v2) represents the probability that entity v1 reaches entity v2 via path P. The probability is then normalized and used to calculate the similarity between two pairs of entities along this path.

[0041] According to one embodiment of the present invention, the prediction score acquisition module 102 is configured to analyze comments on the video using a comment-aware algorithm, extract multidimensional features including content attributes and sentiment attributes, and perform a prediction score on the video to obtain a prediction score.

[0042] Specifically, using the SentiRec review perception algorithm, in the first stage, the user review vector is obtained. The collected user reviews are truncated into sentences of length d. The review set is denoted as R and used as the input of the CNN. The convolution kernel is set to k and the window size is set to s. After extracting features from the review set R, it is sent to the pooling layer to obtain the pooled feature c. C is input into the first fully connected layer to obtain the review vector V, which is then sent to the second fully connected layer to obtain the predicted score. The loss function is:

[0043]

[0044] Then, according to the SentiRec algorithm, in the second stage, the comment vectors are obtained and the ratings are predicted. The vectorized comments are combined into two categories: the user's comment vectors for all video contents V u And the video content receives all users’ comments vector V i , the two types of video vector matrices are input into two CNNs respectively, and after convolution, pooling, and full connection, the user feature F is obtainedu and product feature F i , the inner product of this feature is used to obtain the prediction score r based on user review perception.

[0045] According to one embodiment of the present invention, the recommendation model generation module 103 is configured to fuse the similarity matrix based on content extraction and the similarity matrix based on knowledge reasoning, and use the prediction score obtained by comment perception as an influencing factor to obtain the final recommendation model.

[0046] Specifically, the following formula (5) is used to fuse the similarity matrix based on content extraction and the similarity matrix based on knowledge reasoning through a fusion factor (i.e., the comment perception factor α below). The ratio of the predicted score r to the upper limit of the score is used as the comment perception factor α for enhancement. μ is set to 0.3 according to the heuristic rule to obtain a recommendation model for Top N (first N) video recommendation.

[0047] sim=α·(μ·sim i,text +(1-μ)·sim i,PTranE ) Formula (5)

[0048] Figure 2 A flowchart of a method 200 for video recommendation based on content extraction and rating prediction according to an embodiment of the present invention is shown.

[0049] In step 201, the text content of the target video is extracted to generate a keyword set for the video content. According to one embodiment of the present invention, for video content lacking media resource information, keyframe extraction technology is first used to extract keyframes from the video content. Then, OCR technology is used to identify the text content of the keyframes, and then the corresponding video content description is added. The Jieba word segmentation tool is used to segment the description information of each video. The TF-IDF algorithm is used to calculate the weight of the segmented corpus and extract keywords to form a keyword set for the video content.

[0050] In step 202, a content-based similarity matrix is ​​constructed using a text similarity algorithm based on the keyword set of the video content. According to one embodiment of the present invention, the text descriptions of the videos are analyzed using a TF-IDF or TextRank algorithm, the videos are converted into text vectors, and a text similarity algorithm (e.g., a cosine similarity algorithm) is used to calculate the similarity matrix between the videos.

[0051] In step 203, a knowledge graph is constructed based on the keyword set of the video content, and knowledge reasoning is performed based on the constructed knowledge graph through a path-based translation model to construct a similarity matrix based on knowledge reasoning. According to one embodiment of the present invention, entity extraction is performed on the text content to construct a knowledge graph, and path-based relational reasoning is performed on the video entities in the knowledge graph based on the PTransE algorithm to obtain a video content similarity matrix based on knowledge reasoning. Specifically, using the PTransE data model, the PRA path sorting algorithm is used between the relational entities of the knowledge graph to discover the existence paths between non-adjacent entities. During the specific model training, in order to reduce duplicate data, the length of the random walk is set to l, and new triples with a score of not less than 0.05 are added to the entity set, thereby calculating the similarity between the video contents based on knowledge reasoning.

[0052] Therefore, through step 203, knowledge reasoning methods can be used to supplement the relevance of content, and the PtransE path-based translation model can be used to perform knowledge reasoning on entities in the knowledge graph to expand entity relationships, supplement more similarity measurements between content entities, and give semantics to the video content recommendation process.

[0053] In step 204, a comment-aware algorithm is used to analyze the video comments, extracting multidimensional features including content attributes and sentiment attributes, and predicting the video score to obtain a predicted score. According to one embodiment of the present invention, step 204 includes a first stage and a second stage. In the first stage, user comment vectors are obtained. In the second stage, the comment vectors are obtained and the score is predicted.

[0054] Specifically, in the first stage, video comments are used as a dataset. First, the sentences are preprocessed, and the SentiRec algorithm is used to vectorize the comments through CNN. The predicted score r is calculated, and the comment vector V is obtained by iteration according to the loss function. In the second stage, the comment vectors are spliced ​​into comment vectors V based on the video content. u , the video content receives all users’ comments vector V i , and then input into two CNNs respectively, and after convolution, pooling, and full connection, the user feature F is obtained u and product feature F i , the inner product of this feature is used to obtain the prediction score r based on user review perception.

[0055] In step 205, a fusion factor is used to fuse the content-extracted similarity matrix constructed in step 202 and the knowledge-inference-based similarity matrix constructed in step 203 to obtain the final recommendation model. According to one embodiment of the present invention, the fusion factor used for fusion is based on the review-perception prediction score r obtained in step 204. Specifically, the fusion factor can be the ratio of the prediction score r to the upper limit of the rating. For example, if the prediction score r is 4 and the upper limit of the rating is 5, the fusion factor can be 4 / 5.

[0056] Therefore, the present invention uses the calculation rules of a fusion factor to fuse multiple content-based similarity models and introduces subjective features other than content as enhancement factors to improve the effectiveness of recommendations. Specifically, the present invention uses the fusion factor to combine the results of text extraction and knowledge reasoning to provide a fusion recommendation model, improving the similarity judgment between video content and making the recommendation results more interpretable. In addition, the SentiRec comment perception algorithm is used to automatically extract feature vectors that include both content and emotion, further predicting video ratings. The ratings are used as an enhancement factor for the final recommendation results, improving user acceptance of the recommendations.

[0057] Compared with the prior art, the present invention has the following advantages:

[0058] 1. In existing technologies, some recommendation algorithms construct knowledge graphs, but these graphs are only used to extract meta-information for content association, resulting in a general association model based on the attributes of the content itself:

[0059] This paper uses knowledge reasoning to supplement the relevance of content. It adopts the PtransE path-based translation model to perform knowledge reasoning on non-adjacent entities in the knowledge graph to expand entity relationships, supplement the similarity measurement between different contents, and provide explainability in the video content recommendation process.

[0060] 2. In existing technologies, there are recommendation algorithms that are solely based on historical behavior. However, due to the cold start problem, the recommendation effect is poor:

[0061] The present invention uses the SentiRec algorithm based on convolutional neural networks to automatically extract comment vectors with both content features and emotional expression features from user comments, and further predict video ratings based on comment perception.

[0062] 3. In existing technologies, algorithms that use comments as recommendation information often extract video-related content tags through text analysis as user preference factors for recommendation calculations between different videos, while ignoring the emotional expressions of users during the evaluation process:

[0063] This paper combines a similarity model based on text extraction and knowledge reasoning to improve the semantics of content recommendations. At the same time, it uses the user review scores obtained based on the SentiRec algorithm as an influencing factor to recommend results with better quality evaluation based on content similarity, thereby improving the effectiveness of recommendations.

[0064] Figure 3 A block diagram 300 of an exemplary computing device according to one embodiment of the present invention is shown, which is an example of a hardware device applicable to various aspects of the present invention. The computing device 300 can be any machine that can be configured to perform processing and / or computing, and can be, but is not limited to, a workstation, server, desktop computer, laptop computer, tablet computer, personal digital assistant, smartphone, in-vehicle computer, or any combination thereof. The computing device 300 may include components that can connect or communicate via one or more interfaces and a bus 302. For example, the computing device 300 may include a bus 302, one or more processors 304, one or more input devices 306, and one or more output devices 308. The one or more processors 304 can be any type of processor and can include, but are not limited to, one or more general-purpose processors and / or one or more special-purpose processors (e.g., specialized processing chips). The input device 306 can be any type of device capable of inputting information to the computing device and can include, but are not limited to, a mouse, keyboard, touch screen, microphone, and / or remote control. The output device 308 can be any type of device capable of presenting information and can include, but are not limited to, a display, speakers, video / audio output terminals, vibrators, and / or printers. The computing device 300 may also include or be connected to a non-transient storage device 310, which may be any storage device that is non-transient and capable of storing data, and may include, but is not limited to, a disk drive, an optical storage device, a solid-state memory, a floppy disk, a floppy disk, a hard disk, a magnetic tape or any other magnetic medium, an optical disk or any other optical medium, a ROM (read-only memory), a RAM (random access memory), a cache memory, and / or any memory chip or cassette, and / or any other medium from which a computer can read data, instructions, and / or code. The non-transient storage device 310 may be detachable from the interface. The non-transient storage device 310 may have data / instructions / code for implementing the above-described methods and steps. The computing device 300 may also include a communication device 312. The communication device 312 can be any type of device or system that can communicate with internal devices and / or with a network and can include but is not limited to a modem, a network card, an infrared communication device, a wireless communication device and / or a chipset, such as a Bluetooth device, an IEEE1302.11 device, a WiFi device, a WiMax device, a cellular communication device and / or the like.

[0065] The bus 302 may include, but is not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MCA) bus, an Enhanced ISA (EISA) bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0066] The computing device 300 may also include a working memory 314 , which may be any type of working memory capable of storing instructions and / or data that facilitate the operation of the processor 304 and may include, but is not limited to, random access memory and / or read-only storage devices.

[0067] Software components may be located in the working memory 314, including but not limited to an operating system 316, one or more application programs 318, drivers, and / or other data and code. Instructions for implementing the above-described methods and steps of the present invention may be included in the one or more application programs 318, and the processor 304 may read and execute the instructions of the one or more application programs 318 to implement the above-described method 200 of the present invention.

[0068] It should also be appreciated that variations may be made based on specific needs. For example, custom hardware may also be used, and / or specific components may be implemented in hardware, software, firmware, middleware, microcode, hardware description language, or any combination thereof. In addition, connections to other computing devices, such as network input / output devices, etc., may be employed. For example, some or all of the disclosed methods and apparatus may be implemented using logic and algorithms according to the present invention using programming hardware (e.g., programmable logic circuits including field programmable gate arrays (FPGAs) and / or programmable logic arrays (PLAs)) in assembly language or a hardware programming language (e.g., VERILOG, VHDL, C++).

[0069] Although various aspects of the present invention have been described so far with reference to the accompanying drawings, the above-described methods and apparatus are merely illustrative, and the scope of the present invention is not limited to these aspects, but is defined solely by the appended claims and their equivalents. Various components may be omitted or replaced by equivalent components. In addition, the steps described may be performed in an order different from that described in the present invention. Furthermore, the various components may be combined in various ways. It is also important to note that as technology advances, many of the components described may be replaced by equivalent components that become available later.

Claims

1. A video recommendation method based on content extraction and rating prediction, comprising: Extracting the text content of the video to generate a set of keywords for the video content; Based on the keyword set of the video content, a similarity matrix based on content extraction is constructed using a text similarity algorithm; Based on the keyword set of the video content, a knowledge graph is constructed, and knowledge reasoning is performed on the knowledge graph through a path-based translation model to construct a similarity matrix based on knowledge reasoning; Use a comment-aware algorithm to analyze video comments, extract multidimensional features including content attributes and sentiment attributes, and predict the video to obtain a prediction score. as well as fusing the similarity matrix constructed based on content extraction and the similarity matrix constructed based on knowledge reasoning through a fusion factor to obtain a final recommendation model, wherein the fusion factor is based on the prediction score; The multi-dimensional features include the user's comment vectors on all video contents and the comment vectors received by all users on the video contents.

2. The method according to claim 1, wherein The keyword set for generating video content further includes: Jieba word segmentation tool is used to segment the description information of each video; Through the TF-IDF algorithm, the weight of the segmented corpus is calculated, and the keywords are extracted to form a keyword set for the video content.

3. The method according to claim 1, wherein The text similarity algorithm is a cosine similarity algorithm.

4. The method according to claim 1, wherein The path-based translation model is the PTransE model.

5. The method according to claim 1, wherein The comment-aware algorithm is the SentiRec algorithm.

6. The method of claim 5, wherein performing a prediction score on the video to obtain a prediction score further comprises: Collect video user comment data; Use the comment data as a dataset and use the SentiRec algorithm to train the model to obtain the comment vector; as well as The review vector is input into the SentiRec algorithm training model to extract features, and the prediction score is obtained by performing the inner product of the features.

7. A video recommendation system based on content extraction and rating prediction, comprising: A video content similarity matrix acquisition module, comprising a content extraction-based similarity matrix acquisition module and a knowledge reasoning-based similarity matrix acquisition module, wherein the video content similarity matrix acquisition module is configured to extract text content of a video to generate a keyword set of the video content, wherein the content extraction-based similarity matrix acquisition module is configured to construct a content extraction-based similarity matrix based on the keyword set of the video content by a text similarity algorithm, and wherein the knowledge reasoning-based similarity matrix acquisition module is configured to construct a knowledge graph based on the keyword set of the video content, and perform knowledge reasoning on the knowledge graph by a path-based translation model to construct a knowledge reasoning-based similarity matrix; A prediction score acquisition module configured to analyze video comments using a comment-aware algorithm, extract multidimensional features including content attributes and sentiment attributes, and predictively score the video to obtain a prediction score; the multidimensional features include a vector of user comments on all video content and a vector of comments received by all users on the video content; as well as A recommendation model generation module is configured to fuse the similarity matrix constructed based on content extraction and the similarity matrix constructed based on knowledge reasoning through a fusion factor to obtain a final recommendation model, wherein the fusion factor is based on the prediction score.

8. The system of claim 7, wherein: The text similarity algorithm is a cosine similarity algorithm, the path-based translation model is a PTransE model and / or the comment-aware type algorithm is a SentiRec algorithm.

9. The system of claim 8, wherein: Predicting and scoring the video to obtain a prediction score further includes: Collect video user comment data; Use the review data as a dataset and use the SentiRec algorithm to train the model to obtain the review vector; and The review vector is input into the SentiRec algorithm training model to extract features, and the prediction score is obtained by performing the inner product of the features.

10. A computing device for video recommendation based on content extraction and rating prediction, comprising: processor; A memory storing instructions, wherein the instructions, when executed by the processor, can execute the method according to any one of claims 1 to 6.

Images