CN117972198A

CN117972198A - Course recommendation method, course recommendation device, electronic equipment and medium

Info

Publication number: CN117972198A
Application number: CN202410086616.9A
Authority: CN
Inventors: 杨鑫; 郭斌; 李洪涛; 张丹洁
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2024-01-22
Filing date: 2024-01-22
Publication date: 2024-05-03

Abstract

The course recommendation method, the course recommendation device, the electronic equipment and the medium can be applied to the technical field of big data and the technical field of artificial intelligence. The method comprises the following steps: acquiring a target course requirement of a target user; performing data processing on the target course demands to obtain target course data; traversing the target course data by using a course content knowledge graph to obtain course content similarity, wherein the course content knowledge graph comprises a plurality of entities, a plurality of edges and importance weights corresponding to the entities, the entities are knowledge points obtained by identifying multi-mode platform course information by using a knowledge point identification model and a knowledge point corresponding model, and the edges are relations among the knowledge points; and recommending courses with the similarity of the course content larger than a preset threshold value to the target user.

Description

Course recommendation method, course recommendation device, electronic equipment and medium

Technical Field

The invention relates to the technical field of big data and artificial intelligence, in particular to a course recommendation method, a course recommendation device, electronic equipment and a course recommendation medium.

Background

With the rapid development of information technology, a learning platform of a financial institution is becoming an important channel for knowledge acquisition and skill improvement. These platforms provide a large amount of learning video covering various aspects of the financial field. However, in the face of such a great deal of course resources, especially in the financial field, a knowledge point is often not limited to a single course, but is throughout a plurality of related courses, and a learner is often required to learn systematically about all the related knowledge points.

At present, a course searching mechanism of a learning platform is mainly based on keyword matching of course names, and the method has obvious limitations: because knowledge points in the financial domain may have a variety of different naming schemes, a single keyword search often has difficulty capturing all relevant curriculum content. Thus, it is difficult to provide a comprehensive, systematic learning path for a learner, relying only on the search of course name keywords.

Disclosure of Invention

In view of the above-mentioned problems, according to a first aspect of the present invention, there is provided a course recommendation method including: acquiring a target course requirement of a target user; performing data processing on the target course demands to obtain target course data; traversing the target course data by using a course content knowledge graph to obtain course content similarity, wherein the course content knowledge graph comprises a plurality of entities, a plurality of edges and importance weights corresponding to the entities, the entities are knowledge points obtained by identifying multi-mode platform course information by using a knowledge point identification model and a knowledge point corresponding model, and the edges are relations among the knowledge points; and recommending courses with the similarity of the course content larger than a preset threshold value to the target user.

According to some exemplary embodiments, the multi-modal platform lesson information includes text information, image information and audio information, and the plurality of entities identify with a knowledge point identification model and a knowledge point correspondence model based on the multi-modal platform lesson information, specifically including: performing first preprocessing on the text information to obtain sequence labeling data divided in sentence level; inputting the sequence labeling data into the knowledge point recognition model, and outputting a first label sequence corresponding to the text information; performing second preprocessing on the image information to obtain image characteristics; performing third preprocessing on the audio information to obtain audio characteristics; inputting the image features and the audio features into the knowledge point corresponding model to perform knowledge point mapping to obtain a second tag sequence and a third tag sequence; and obtaining the plurality of entities based on the first tag sequence, the second tag sequence, and the third tag sequence.

According to some exemplary embodiments, the knowledge point recognition model includes an embedded layer, a convolutional neural network layer, a full-connection layer, a bidirectional long and short-time memory layer and a conditional random field layer which are connected in series, the sequence labeling data is input into the knowledge point recognition model, and a first tag sequence corresponding to text information is output, and the method specifically includes: the embedded layer converts the sequence labeling data into an embedded vector sequence; the convolutional neural network layer extracts candidate entities of the embedded vector sequence; the full connection layer performs dimension reduction on the candidate entity to obtain a dimension reduction vector sequence; the two-way long short-time memory layer extracts sentence level information in the dimension reduction vector sequence and outputs a prediction tag sequence; and the conditional random field layer carries out rule constraint of knowledge point identification on the predicted tag sequence and outputs the first tag sequence.

According to some exemplary embodiments, the convolutional neural network layer sets a plurality of convolutional kernel sizes, and the convolutional neural network layer extracts candidate entities of the embedded vector sequence, specifically including: enumerating all candidate entities using a convolution kernel based on the embedded vector sequence; and encoding the candidate entity to the fully-connected layer through convolution operation.

According to some exemplary embodiments, the relationship between the knowledge points is obtained by using a transducer model, wherein the method specifically comprises the following steps: taking the first tag sequence, the second tag sequence and the third tag sequence as input, and performing relation extraction by using the transducer model to obtain knowledge point relation extraction features; and inputting the knowledge point relation extraction features into an activation function layer to acquire the relation among the knowledge points.

According to some exemplary embodiments, the importance weight is determined based on the frequency of occurrence of the knowledge points.

According to some exemplary embodiments, the data processing for the target course requirement to obtain target course data specifically includes: extracting audio text of the target course requirement and cleaning the text under the condition that the target course requirement is in an audio form; performing text cleaning on the target course requirement under the condition that the target course requirement is in a text form; extracting keywords and phrases from the target course requirement after text cleaning to obtain key information; and converting the format of the key information to obtain the target course data.

According to a second aspect of the present invention, there is provided a course recommendation apparatus, the apparatus comprising: the target course demand acquisition module is used for: acquiring a target course requirement of a target user; a data processing module for: performing data processing on the target course demands to obtain target course data; the course content similarity acquisition module is used for: traversing the target course data by using a course content knowledge graph to obtain course content similarity, wherein the course content knowledge graph comprises a plurality of entities, a plurality of edges and importance weights corresponding to the entities, the entities are knowledge points obtained by identifying multi-mode platform course information by using a knowledge point identification model and a knowledge point corresponding model, and the edges are relations among the knowledge points; and a course recommendation module for: and recommending courses with the similarity of the course content larger than a preset threshold value to the target user.

According to some exemplary embodiments, the data processing module may include a first text cleansing unit, a second text cleansing unit, a key information acquisition unit, and a format conversion unit.

According to some exemplary embodiments, the first text cleansing unit may be configured to extract audio text of the target lesson requirement and conduct text cleansing in case the target lesson requirement is in audio form.

According to some exemplary embodiments, the second text cleansing unit may be configured to perform text cleansing on the target lesson requirement in case the target lesson requirement is in text form.

According to some exemplary embodiments, the key information obtaining unit may be configured to extract keywords and phrases from the target course requirement after text cleaning, to obtain key information.

According to some exemplary embodiments, the format conversion unit may be configured to perform format conversion on the key information to obtain the target course data.

According to some example embodiments, the course content similarity obtaining module may include an entity identification module and a relationship extraction module.

According to some example embodiments, the entity identification module may include a first preprocessing unit, a first tag sequence acquisition module, a second preprocessing unit, a third preprocessing unit, a mapping unit, and an entity acquisition unit.

According to some exemplary embodiments, the first preprocessing unit may be configured to perform a first preprocessing on the text information, to obtain sequence annotation data divided at a sentence level.

According to some exemplary embodiments, the first tag sequence obtaining module may be configured to input the sequence labeling data into the knowledge point recognition model, and output a first tag sequence corresponding to the text information.

According to some exemplary embodiments, the second preprocessing unit may be configured to perform a second preprocessing on the image information to obtain image features.

According to some exemplary embodiments, the third preprocessing unit may be configured to perform third preprocessing on the audio information to obtain audio features.

According to some exemplary embodiments, the mapping unit may be configured to input the image feature and the audio feature into the knowledge point correspondence model to perform knowledge point mapping, to obtain a second tag sequence and a third tag sequence.

According to some example embodiments, the entity obtaining unit may be configured to obtain the plurality of entities based on the first tag sequence, the second tag sequence, and the third tag sequence.

According to some example embodiments, the first tag sequence obtaining module may include an embedded vector sequence converting unit, a candidate entity extracting module, a reduced-dimension vector sequence obtaining unit, a predicted tag sequence obtaining unit, and a first tag sequence obtaining unit.

According to some example embodiments, the embedded vector sequence conversion unit may be configured to convert the sequence annotation data into an embedded vector sequence by the embedding layer.

According to some example embodiments, the candidate entity extraction module may be configured to extract candidate entities of the embedded vector sequence from the convolutional neural network layer.

According to some exemplary embodiments, the dimension-reduction vector sequence obtaining unit may be configured to perform dimension reduction on the candidate entity by using the full connection layer to obtain a dimension-reduction vector sequence.

According to some exemplary embodiments, the predictive tag sequence acquiring unit may be configured to extract sentence-level information in the dimension-reduced vector sequence by the bidirectional long short-time memory layer and output a predictive tag sequence.

According to some exemplary embodiments, the first tag sequence obtaining unit may be configured to output the first tag sequence under a rule constraint that the conditional random field layer performs knowledge point identification on the predicted tag sequence.

According to some example embodiments, the candidate entity extraction module includes an enumeration unit and a convolution unit.

According to some example embodiments, the enumeration unit may be configured to enumerate all candidate entities using a convolution kernel based on the embedded vector sequence.

According to some example embodiments, the convolution unit may be configured to encode the candidate entity to the fully-connected layer by a convolution operation.

According to some example embodiments, the relationship extraction module may include a knowledge point relationship extraction feature acquisition unit and a relationship acquisition unit.

According to some exemplary embodiments, the knowledge point relation extraction feature obtaining unit may be configured to obtain a knowledge point relation extraction feature by using the transform model to perform relation extraction with the first tag sequence, the second tag sequence, and the third tag sequence as input.

According to some exemplary embodiments, the relationship obtaining unit may be configured to input the knowledge point relationship extraction feature to an activation function layer to obtain a relationship between the knowledge points.

According to a third aspect of the present invention, there is provided an electronic device comprising: one or more processors; and a storage device for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the method as described above.

According to a fourth aspect of the present invention there is provided a computer readable storage medium having stored thereon executable instructions which when executed by a processor cause the processor to perform a method as described above.

According to a fifth aspect of the present invention there is provided a computer program product comprising a computer program which, when executed by a processor, implements a method as described above.

One or more of the above embodiments have the following advantages or benefits: according to the course recommendation method provided by the invention, the knowledge points are utilized to identify the multi-mode platform course information to construct the knowledge map, so that the platform course key information can be reflected more accurately from multi-mode multi-dimension, the manual processing requirement is reduced, and the overall processing efficiency is improved; meanwhile, the course content knowledge graph provides structured information of course content, so that data query and processing are more efficient. In addition, by analyzing the specific requirements of the user, highly personalized course recommendation can be provided through the course content knowledge graph, and the correlation and accuracy of recommendation are improved, so that the time and effort of the user when searching for a proper course are reduced, and the user experience is improved.

Drawings

The foregoing and other objects, features and advantages of the invention will be apparent from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

Fig. 1 schematically illustrates an application scenario diagram of a course recommendation method, apparatus, device, medium according to an embodiment of the invention.

FIG. 2 schematically illustrates a flow chart of a course recommendation method according to an embodiment of the invention.

FIG. 3 schematically illustrates a flow chart of a method of data processing a target lesson requirement in accordance with an embodiment of the invention.

Fig. 4 schematically shows a flow chart of a method for entity recognition using a knowledge point recognition model and a knowledge point correspondence model, in accordance with an embodiment of the invention.

Fig. 5 schematically shows a flowchart of a method for outputting a first tag sequence corresponding to text information using a knowledge point recognition model, in accordance with an embodiment of the invention.

Fig. 6 schematically shows a flow chart of a method of extracting candidate entities according to an embodiment of the invention.

Fig. 7 schematically shows a flow chart of a method of obtaining relationships between knowledge points, in accordance with an embodiment of the invention.

FIG. 8 schematically illustrates enumeration using different size convolution kernels according to an embodiment of the present disclosure.

Fig. 9 schematically shows a block diagram of the course recommending apparatus according to the embodiment of the present invention.

Fig. 10 schematically illustrates a block diagram of an electronic device adapted for a course recommendation method according to an embodiment of the invention.

Detailed Description

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. It should be understood that the description is only illustrative and is not intended to limit the scope of the invention. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the invention. It may be evident, however, that one or more embodiments may be practiced without these specific details. In addition, in the following description, descriptions of well-known structures and techniques are omitted so as not to unnecessarily obscure the present invention.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The terms "comprises," "comprising," and/or the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.

All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It should be noted that the terms used herein should be construed to have meanings consistent with the context of the present specification and should not be construed in an idealized or overly formal manner.

Where a convention analogous to "at least one of A, B and C, etc." is used, in general such a convention should be interpreted in accordance with the meaning of one of skill in the art having generally understood the convention (e.g., "a system having at least one of A, B and C" would include, but not be limited to, systems having a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).

In the technical scheme of the invention, the acquisition, storage, application and the like of the related personal information of the user accord with the regulations of related laws and regulations, necessary security measures are taken, and the public order harmony is not violated.

First, technical terms described herein are explained and illustrated as follows.

Knowledge graph is a structured knowledge representation that represents data in the form of graphs, where nodes represent entities (e.g., people, places, organizations, concepts, etc.), and edges represent various relationships between entities. Knowledge maps are widely used in various fields such as search engines, recommendation systems, natural language processing, artificial intelligence, and the like.

The embedded layer (Embedding Layer) is an important concept in the field of machine learning and Natural Language Processing (NLP), especially when processing text data. The primary function of the embedding layer is to convert a large number of discrete symbols (e.g., words, characters, etc.) into dense vectors of fixed size, which are typically real-type and contain some semantic information for each dimension. The purpose of the embedded layer is to capture similarities and differences between symbols and present this information in a manner that is more suitable for machine learning model processing.

The convolutional neural network layer (Convolutional Neural Network Layer, CNN layer for short) is a core concept in deep learning, and is mainly used in the fields of image processing, video analysis, natural language processing and the like. The CNN layer extracts features of input data (e.g., images or text) through a convolution operation. This process involves the use of a set of learnable filters (or convolution kernels) that slide over the input data to capture local features.

Each neuron in the fully connected layer is connected to all neurons of the previous layer. This means that every incoming message in the network affects the output of the fully connected layer. The main purpose of the fully connected layer is to integrate the features extracted from the previous layer (e.g., convolutional layer or pooling layer) and ultimately for classification or other tasks.

The bidirectional long short term memory layer (Bidirectional Long Short-Term Memory Layer, biLSTM) is a special type of Recurrent Neural Network (RNN) layer that is commonly used to process sequence data, such as text or time series. LSTM (long and short term memory) is a special RNN designed to solve the problem of gradient extinction or gradient explosion of a conventional RNN when processing long sequence data. The bi-directional LSTM (BiLSTM) extends on this basis to enhance the model's understanding of the sequence data by combining information in both directions. It contains both forward (from past to future) and reverse (from never to past) LSTM, processing sequence data separately.

The conditional random field layer (Conditional Random FIELD LAYER, abbreviated CRF layer) is a structured predictive model for modeling sequence data to model the conditional dependencies between nodes (e.g., words) and labels (e.g., solid types) observed in the sequence data. In the CRF layer, the model considers not only the optimal label corresponding to each node alone, but also the dependency relationship between adjacent labels, thereby performing globally optimal sequence prediction.

The transducer model is a deep learning architecture, and the key characteristic is that a self-attention mechanism is used. This mechanism enables the model to capture relationships between all elements in a sequence while processing it.

Under the push of the digital age, financial education and training has rapidly transformed to online platforms, which provide rich and diverse course resources, and meet diversified demands of practitioners and students on financial knowledge. However, while online learning platforms have provided unprecedented convenience to users, a number of challenges have arisen, particularly in the organization and recommendation of curriculum content.

Currently, the course search and recommendation functions of most learning platforms are mainly based on keyword matching, which often appears insufficient for financial learning, especially for advanced expertise mastering. Knowledge points and concepts in the financial field are highly specialized and complex, a concept may be expressed in multiple ways, and a topic may span multiple courses. Therefore, it is difficult to comprehensively cover all courses related to a specific knowledge point based on a single or simple keyword search, resulting in fragmentation of a learning path and incomplete knowledge grasping.

In addition, financial education is characterized by a tight combination of theory and practice, and the knowledge points involved often require interleaving learning and practical application among multiple courses. Therefore, a single course is difficult to meet the demands of students on deep and systematic learning, and the students want to learn a piece of knowledge systematically, so that only one course is insufficient, but the students need to learn all courses related to the related knowledge, so that the knowledge can be comprehensively known, and the students need to learn systematically for a large number of courses of related knowledge points. There is a need for a method that can accurately find relevant courses through knowledge venues to be recommended to users based on course relevance.

Based on this, an embodiment of the present invention provides a course recommendation method, which includes: acquiring a target course requirement of a target user; performing data processing on the target course demands to obtain target course data; traversing the target course data by using a course content knowledge graph to obtain course content similarity, wherein the course content knowledge graph comprises a plurality of entities, a plurality of edges and importance weights corresponding to the entities, the entities are knowledge points obtained by identifying multi-mode platform course information by using a knowledge point identification model and a knowledge point corresponding model, and the edges are relations among the knowledge points; and recommending courses with the similarity of the course content larger than a preset threshold value to the target user. According to the course recommendation method provided by the invention, the knowledge points are utilized to identify the multi-mode platform course information to construct the knowledge map, so that the platform course key information can be reflected more accurately from multi-mode multi-dimension, the manual processing requirement is reduced, and the overall processing efficiency is improved; meanwhile, the course content knowledge graph provides structured information of course content, so that data query and processing are more efficient. In addition, by analyzing the specific requirements of the user, highly personalized course recommendation can be provided through the course content knowledge graph, and the correlation and accuracy of recommendation are improved, so that the time and effort of the user when searching for a proper course are reduced, and the user experience is improved.

It should be noted that the course recommendation method, device, equipment and medium determined by the present invention can be used in the big data technical field and the artificial intelligence technical field, and also can be used in the financial field, and can be used in various fields other than the big data technical field and the artificial intelligence technical field as well as the financial field. The application fields of the course recommendation method, the device, the equipment and the medium provided by the embodiment of the invention are not limited.

In the technical scheme of the invention, the related user information (including but not limited to user personal information, user image information, user equipment information, such as position information and the like) and data (including but not limited to data for analysis, stored data, displayed data and the like) are information and data authorized by a user or fully authorized by all parties, and the processing of the related data such as collection, storage, use, processing, transmission, provision, disclosure, application and the like are all conducted according to the related laws and regulations and standards of related countries and regions, necessary security measures are adopted, no prejudice to the public welfare is provided, and corresponding operation inlets are provided for the user to select authorization or rejection.

As shown in fig. 1, an application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. Various communication client applications, such as shopping class applications, web browser applications, search class applications, instant messaging tools, mailbox clients, social platform software, etc. (by way of example only) may be installed on the terminal devices 101, 102, 103.

The terminal devices 101, 102, 103 may be a variety of electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablets, laptop and desktop computers, and the like.

The server 105 may be a server providing various services, such as a background management server (by way of example only) providing support for websites browsed by users using the terminal devices 101, 102, 103. The background management server may analyze and process the received data such as the user request, and feed back the processing result (e.g., the web page, information, or data obtained or generated according to the user request) to the terminal device.

It should be noted that the course recommendation method provided by the embodiment of the present invention may be generally performed by the server 105. Accordingly, the course recommendation device provided in the embodiment of the present invention may be generally disposed in the server 105. The course recommendation method provided by the embodiment of the present invention may also be performed by a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105. Accordingly, the course recommendation apparatus provided in the embodiment of the present invention may also be provided in a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105.

It should be understood that the number of devices and networks in fig. 1 is merely illustrative. There may be any number of devices and networks, as desired for implementation.

As shown in fig. 2, the course recommendation method 200 of this embodiment may include operations S210 to S240.

In operation S210, a target course demand of a target user is acquired.

In an embodiment of the invention, users may describe their learning needs by entering a piece of text; the text description may include the subject matter of the lesson, a specific knowledge point, a difficulty level of learning, or any other specific requirement. In addition, for users who prefer to verbally express or seek a more convenient way, their learning needs can also be expressed by speech descriptions.

In operation S220, data processing is performed on the target course requirement to obtain target course data.

As shown in fig. 3, the method for performing data processing on a target course requirement according to the embodiment may include operations S310 to S340, and operations S310 to S340 may at least partially perform operation S220.

In operation S310, in case that the target course requirement is in an audio form, audio text of the target course requirement is extracted and text washing is performed.

In an embodiment of the present invention, if the user's lesson needs are presented in audio form, it is necessary to extract text content in the audio. Specifically, this may be accomplished by an automatic speech recognition system. After the audio text is extracted, text cleaning can be performed to remove irrelevant noise, non-language elements (such as cough sound, background noise and the like) and language redundancy (such as repeated vocabulary, spoken Buddhist and the like), and to remove special symbols, punctuation, non-standard vocabulary and the like in the text. Text cleansing may also involve a unified lexical format to reduce the complexity of subsequent processing.

In operation S320, in the case that the target course requirement is in a text form, text cleansing is performed on the target course requirement.

In an embodiment of the present invention, if the user' S course requirements are directly presented in text form, the target course requirements may be text-purged directly according to the procedure involving text purging in operation S310.

In operation S330, keyword and phrase extraction is performed on the target course requirement after text washing, and key information is obtained.

In embodiments of the present invention, keyword and phrase extraction may be performed using natural language processing techniques such as TF-IDF analysis, LDA topic modeling, or other advanced text analysis methods to obtain key information. The extraction of keywords and phrases helps capture core content required by the user, such as specific lesson topics, knowledge points, or skills.

In operation S340, format conversion is performed on the key information to obtain the target course data.

In embodiments of the present invention, the extracted key information may be converted into a format suitable for subsequent processing. For example, it may include converting keywords and phrases to specific codes or tags to match course content knowledge maps.

Referring back to fig. 2, in operation S230, the target course data is traversed by using a course content knowledge graph to obtain a course content similarity, where the course content knowledge graph includes a plurality of entities, a plurality of edges, and importance weights corresponding to the plurality of entities, the plurality of entities are knowledge points obtained by identifying multi-modal platform course information by using a knowledge point identification model and a knowledge point correspondence model, and the plurality of edges are relationships between the knowledge points.

In an embodiment of the present invention, the multimodal platform lesson information may include text information, image information, and audio information. The text information may relate to subtitles, course introduction, ppt text content, lectures, etc. appearing in the platform course.

In embodiments of the present invention, the image information and audio information may be from video in a platform lesson, and thus the process of obtaining these two modality data typically involves parsing and processing the video file. Specifically, the method comprises the following steps: the video file may be parsed using video processing software or a programming library and audio and image data streams extracted therefrom; extracting audio tracks from video files, which is typically a relatively straightforward process, may result in a single audio file or data stream; the extracted audio may be further processed, such as denoising, volume normalization, speech recognition, etc., to extract useful information.

In embodiments of the present invention, the extraction of image information further involves decomposing the video into a series of frames, each frame being an independent image that can be analyzed separately; the particular frames (e.g., the first frame of each scene or periodically sampled frames) may be selected for analysis as desired, which helps reduce the amount of processing data while preserving critical visual information. The method integrates the extraction and analysis of the audio and image information, can acquire rich multi-mode data from the video, and is very helpful for understanding the video content and improving the accuracy and depth of course recommendation.

As shown in fig. 4, the method for entity recognition using the knowledge point recognition model and the knowledge point correspondence model of this embodiment may include operations S410 to S460, and operations S410 to S460 may at least partially perform operation S230.

In operation S410, the text information is subjected to a first preprocessing to obtain sequence annotation data divided at sentence level.

In embodiments of the present invention, the first preprocessing may include text cleansing, word segmentation, and segmentation of the complete text into smaller units, i.e., sentence-level. Specifically, the text may be divided in sentence units, each sentence is separated by a blank line, one sentence is a piece of sample data, and each Chinese character in the sentence is separately in a line.

In operation S420, the sequence labeling data is input into the knowledge point recognition model, and a first tag sequence corresponding to the text information is output.

In the embodiment of the invention, the knowledge point identification model comprises an embedded layer, a convolutional neural network layer, a full-connection layer, a two-way long and short-term memory layer and a conditional random field layer which are connected in series.

As shown in fig. 5, the method of outputting the first tag sequence corresponding to the text information using the knowledge point recognition model of this embodiment may include operations S510 to S550, and operations S510 to S550 may at least partially perform operation S420.

In operation S510, the embedding layer converts the sequence annotation data into an embedded vector sequence.

In an embodiment of the present invention, the embedding layer may convert the input sequence annotation data into an embedded vector sequence. Each character is mapped to a vector in a high-dimensional space that captures the semantic features of the vocabulary.

In operation S520, the convolutional neural network layer extracts candidate entities of the embedded vector sequence.

In an embodiment of the invention, the convolutional neural network layer may receive vector sequences from the embedded layer and process the sequences through a plurality of convolutional checks. Each convolution kernel is responsible for capturing specific features or patterns, such as word context information. Convolution kernels of multiple sizes can cover combinations of words of different lengths, helping to extract entities of different sizes and enhancing the understanding of the model to the text.

FIG. 6 schematically illustrates a flow chart of a method of extracting candidate entities according to an embodiment of the invention; FIG. 8 schematically illustrates enumeration using different size convolution kernels according to an embodiment of the present disclosure.

As shown in fig. 6, the method for extracting candidate entities of this embodiment may include operations S610 to S620, and operations S610 to S620 may at least partially perform operation S520.

In operation S610, all candidate entities are enumerated using a convolution kernel based on the embedded vector sequence.

In embodiments of the present invention, a single-channel two-dimensional convolution kernel may be employed to traverse the entire embedded vector sequence to identify and enumerate all possible candidate entities, which may be specific words, phrases, or any meaningful text segments. A variety of different sizes of convolution kernels can capture entities of different lengths.

As shown in fig. 8, a smaller convolution kernel (shown as a dashed box) may capture an entity of a single or short word, while a larger convolution kernel (shown as a solid box) may capture a longer phrase or continuous word sequence, i.e., information of "site employee" and "employee" in "how to excite the work enthusiasm of site employees" is extracted by the model, not just selected from "site", "site person" and "employee". This process relies on the local perceptibility of convolutional neural networks to effectively identify structured information from complex text data.

In operation S620, the candidate entity is encoded to the fully-connected layer through a convolution operation.

In embodiments of the present invention, once candidate entities are identified, these entities may be encoded by a convolution operation into a format that the fully connected layer can handle. In this process, the convolution operation not only extracts the features of the entity, but also converts those features into a more dense and meaningful representation, which may include contextual information, semantic characteristics, etc. of the entity.

Referring back to fig. 5, in operation S530, the full connection layer performs dimension reduction on the candidate entity to obtain a dimension-reduced vector sequence.

In the embodiment of the invention, the full connection layer can perform data dimension reduction on the candidate entity output by the multi-size convolution kernel layer, filter noise in the candidate entity, and output a vector sequence after dimension reduction.

In operation S540, the bidirectional long short-term memory layer extracts sentence-level information in the reduced-dimension vector sequence and outputs a predictive tag sequence.

In the embodiment of the invention, the bidirectional long short-time memory layer can extract the global feature of the text, namely sentence level information, from the dimension-reduction vector sequence output by the full-connection layer and forecast the label sequence corresponding to the sequence labeling data of the input model.

In operation S550, the conditional random field layer performs rule constraint of knowledge point recognition on the predicted tag sequence, and outputs the first tag sequence.

In the embodiment of the invention, the conditional random field layer can be used for applying a certain rule constraint for identifying a named entity to the predicted tag sequence output by the bidirectional long and short time memory layer and decoding and outputting a final first tag sequence.

Referring back to fig. 4, in operation S430, the image information is subjected to a second preprocessing to obtain image features.

In the embodiment of the invention, the extracted image frame can be further processed, such as noise reduction, RGB color space to HSV color space conversion, RGB color space to HSI color space conversion, extraction of graphic features such as color histogram, color moment, color aggregate vector and the like, on the input image, so that the extracted image frame is suitable for the subsequent analysis step.

In operation S440, a third preprocessing is performed on the audio information to obtain audio features.

In the embodiment of the invention, operations such as denoising, pre-emphasis, windowing and framing of audio information can be performed, so that the signal quality is improved, and the high-frequency part of the signal is flat and the whole is stable. Endpoint detection and MFCC coefficient extraction operations may also be performed to extract parameters that may represent the essential characteristics of the speech signal.

In operation S450, the image feature and the audio feature are input into the knowledge point corresponding model to perform knowledge point mapping, so as to obtain a second tag sequence and a third tag sequence.

In the embodiment of the invention, the corresponding relation between the extracted image and audio features and the knowledge points can be trained, a knowledge point corresponding model is established, and model parameters are stored in a system, so that the knowledge point mapping is carried out on the input image features and audio features by using the knowledge point corresponding model. In particular, a deep learning model may be created having a plurality of input branches, each input branch receiving a different modality data; each input branch may contain a feature extractor of a particular modality, such as Convolutional Neural Network (CNN) for images, recurrent Neural Network (RNN) for audio; in each branch, features are extracted and passed into a shared hidden layer. Finally, the outputs of all branches are combined and connected to the output layer for mapping and classification of knowledge points.

It should be noted that the above method of creating a deep learning model is only an example, and is not intended to limit the method of obtaining the knowledge point corresponding model. Those skilled in the art will appreciate that the acquisition of the knowledge point correspondence model may also be performed by other methods.

In operation S460, the plurality of entities are obtained based on the first tag sequence, the second tag sequence, and the third tag sequence.

In addition, in order to further construct the knowledge graph, the embodiment of the invention also provides a method for extracting the relationship among the knowledge points.

As shown in fig. 7, the method of acquiring the relationship between knowledge points of this embodiment may include operations S710 to S720, and operations S710 to S720 may at least partially perform operation S230.

In operation S710, the relationship extraction is performed by using the transducer model with the first tag sequence, the second tag sequence, and the third tag sequence as inputs, and knowledge point relationship extraction features are obtained.

In embodiments of the present invention, the three tag sequences entered may represent different information or entities, such as knowledge points, entities, keywords, etc. The use of a transfomer model may allow the model to dynamically focus on the information of other tags as each tag is processed, which helps capture semantic associations between tags. Multiple attention heads may be included in the transducer model, each of which may learn to focus on different semantic information. This helps to improve the characterizability of the model. In each layer of the transducer model, features are extracted and encoded, including semantic information, associations, etc. of tag sequences. Finally, after processing through the transducer model, relationship extraction features are derived that represent relationships, correlations, and/or other information between tag sequences.

In operation S720, the knowledge point relation extraction feature is input to an activation function layer, and the relation between the knowledge points is obtained.

In embodiments of the invention, the activation function may be used to further process the features to obtain relationships between knowledge points, which may include non-linear transformations, normalization, or other operations on the features.

In the embodiment of the invention, the knowledge point recognition model and the knowledge point corresponding model are used for carrying out relation extraction on the knowledge points obtained by recognizing the multi-mode platform course information and the transducer model, so that a knowledge relation triplet can be obtained.

In the embodiment of the invention, the importance weight can be determined based on the occurrence frequency of the knowledge points so as to obtain a knowledge relation quadruple.

In the embodiment of the invention, the importance of the knowledge points can be defined in frequency, and the knowledge points are numbered from 1 to n according to the importance degree. Therefore, an importance parameter can be assigned to each knowledge point, and the importance parameters of the sequence numbers 1 to n are respectively n to 1. For example, knowledge points labeled with sequence numbers 1 through n may be written as 1:2:3:4:5 is divided into 1 to 5 grades, importance parameters of knowledge points are increased according to the knowledge point relation triples, the importance parameters of 5 to 1 are respectively added on the basis of the original knowledge points of 1 to 5 grades, for example, the importance parameters of the knowledge points 1 are n, triples related to the knowledge points 1 are found, the knowledge points 1 and the knowledge points 2 are found to have a relation in the found triples, the importance parameters of the knowledge points 1 are added with 5 on the basis of n, the knowledge points are ranked again according to the importance parameters, and the steps are repeatedly executed until the sequence number of the knowledge points is unchanged. Thus, weights can be added on the basis of the triples to identify the importance degree of the knowledge relationship, and the quadruples are formed.

In the embodiment of the invention, the obtained quadruple can be stored in a graph database, an alignment algorithm is written, the threshold value is set to be a fixed value, for example, 70%, and entity fusion can be performed if the similarity is greater than 70%.

Referring back to fig. 2, in operation S240, a course with the similarity of course content greater than a preset threshold is recommended to the target user.

In the embodiment of the invention, the similarity percentage can be obtained by comparing the weights of the knowledge point level and all the four-element groups where the knowledge point is located with the similarity of the frame relationship. According to the calculated course similarity, courses with similarity higher than a preset threshold value, for example 80%, which are considered to have high correlation with the demands of users, can be screened out. Finally, the selected relevant courses may be recommended to the target user, and the courses may be ranked according to the similarity so that the user may see that the most relevant course is recommended first.

According to the course recommendation method provided by the invention, the knowledge graph constructed by identifying the multi-mode platform course information by using the knowledge point identification model and the knowledge point corresponding model can reflect the platform course key information more accurately from multi-mode multi-dimension, so that the manual processing requirement is reduced, and the overall processing efficiency is improved; meanwhile, the course content knowledge graph provides structured information of course content, so that data query and processing are more efficient. In addition, by analyzing the specific requirements of the user, highly personalized course recommendation can be provided through the course content knowledge graph, and the correlation and accuracy of recommendation are improved, so that the time and effort of the user when searching for a proper course are reduced, and the user experience is improved. Specifically, the following beneficial effects are brought:

1. by processing the target course requirements, the system can more efficiently process the user-provided requirement text or voice description to convert it into target course data, which can reduce processing time and computing resource requirements;

2. Traversing the target course data by using the course content knowledge graph, and enabling the system to rapidly position related course content according to the entity, side and weight information in the graph, so that the course similarity calculation and course recommendation process can be accelerated;

3. by screening courses with similarity larger than a preset threshold value from the knowledge graph, the system can generate a course recommendation list more rapidly so as to respond to the demands of users;

4. due to the adoption of the calculation method of the similarity of course content, users can obtain course recommendations which are more matched with the requirements of the users, so that the user satisfaction is improved, and learning resources which meet the requirements of the users can be more easily found by the users;

5 users can more fully know the specific knowledge field through the recommended related courses, not only limited to a single course, which is helpful for improving the learning effect and knowledge breadth of the users;

6, the user does not need to search and screen related courses manually, and the system can automatically provide the most relevant recommendation, so that the time and energy of the user are saved;

7. By generating personalized course recommendations based on the user's target course needs, the user experience will be more personalized and satisfied.

Based on the course recommendation method, the invention further provides a course recommendation device. The device will be described in detail below in connection with fig. 9.

As shown in fig. 9, the course recommendation apparatus 900 according to the embodiment includes a target course demand acquisition module 910, a data processing module 920, a course content similarity acquisition module 930, and a course recommendation module 940.

The target curriculum requirements retrieval module 910 can be configured to retrieve target curriculum requirements of a target user. In an embodiment, the target course requirement obtaining module 910 may be configured to perform the operation S210 described above, which is not described herein.

The data processing module 920 may be configured to perform data processing on the target course requirement to obtain target course data. In an embodiment, the data processing module 920 may be configured to perform the operation S220 described above, which is not described herein.

The course content similarity obtaining module 930 may be configured to traverse the target course data by using a course content knowledge graph to obtain a course content similarity, where the course content knowledge graph includes a plurality of entities, a plurality of edges, and importance weights corresponding to the plurality of entities, the plurality of entities are knowledge points obtained by identifying multi-mode platform course information by using a knowledge point identification model and a knowledge point correspondence model, and the plurality of edges are relationships between the knowledge points. In an embodiment, the course content similarity obtaining module 930 may be configured to perform the operation S230 described above, which is not described herein.

The course recommendation module 940 may be configured to recommend courses with the similarity of the course content greater than a preset threshold to the target user. In an embodiment, the course recommendation module 940 may be configured to perform the operation S240 described above, which is not described herein.

According to an embodiment of the present invention, the data processing module 920 may include a first text cleansing unit, a second text cleansing unit, a key information acquiring unit, and a format converting unit.

The first text cleansing unit may be configured to extract an audio text of the target lesson demand and conduct text cleansing in a case where the target lesson demand is in audio form. In an embodiment, the first text cleansing unit may be used to perform the operation S310 described above, which is not described herein.

The second text cleansing unit may be configured to conduct text cleansing on the target lesson requirement in case the target lesson requirement is in text form. In an embodiment, the second text cleansing unit may be used to perform the operation S320 described above, which is not described herein.

The key information acquisition unit can be used for extracting keywords and phrases from the target course requirement after text cleaning to acquire key information. In an embodiment, the key information obtaining unit may be configured to perform the operation S330 described above, which is not described herein.

The format conversion unit may be configured to perform format conversion on the key information to obtain the target course data. In an embodiment, the format conversion unit may be configured to perform the operation S340 described above, which is not described herein.

According to an embodiment of the present invention, the course content similarity obtaining module 930 may include an entity identification module and a relationship extraction module.

According to an embodiment of the present invention, the entity identification module may include a first preprocessing unit, a first tag sequence acquisition module, a second preprocessing unit, a third preprocessing unit, a mapping unit, and an entity acquisition unit.

The first preprocessing unit may be configured to perform first preprocessing on the text information, and obtain sequence labeling data divided in sentence level. In an embodiment, the first preprocessing unit may be configured to perform the operation S410 described above, which is not described herein.

The first tag sequence obtaining module may be configured to input the sequence labeling data into the knowledge point recognition model, and output a first tag sequence corresponding to the text information. In an embodiment, the first tag sequence obtaining module may be configured to perform the operation S420 described above, which is not described herein.

The second preprocessing unit may be configured to perform a second preprocessing on the image information to obtain an image feature. In an embodiment, the second preprocessing unit may be configured to perform the operation S430 described above, which is not described herein.

The third preprocessing unit may be configured to perform third preprocessing on the audio information to obtain audio features. In an embodiment, the third preprocessing unit may be configured to perform the operation S440 described above, which is not described herein.

The mapping unit may be configured to input the image feature and the audio feature into the knowledge point corresponding model to perform knowledge point mapping, so as to obtain a second tag sequence and a third tag sequence. In an embodiment, the mapping unit may be configured to perform the operation S450 described above, which is not described herein.

The entity obtaining unit may be configured to obtain the plurality of entities based on the first tag sequence, the second tag sequence, and the third tag sequence. In an embodiment, the entity obtaining unit may be configured to perform the operation S460 described above, which is not described herein.

According to an embodiment of the present invention, the first tag sequence obtaining module may include an embedded vector sequence converting unit, a candidate entity extracting module, a dimension-reduced vector sequence obtaining unit, a predicted tag sequence obtaining unit, and a first tag sequence obtaining unit.

The embedded vector sequence conversion unit may be configured to convert the sequence annotation data into an embedded vector sequence by the embedding layer. In an embodiment, the embedded vector sequence conversion unit may be configured to perform the operation S510 described above, which is not described herein.

The candidate entity extraction module may be configured to extract candidate entities of the embedded vector sequence from the convolutional neural network layer. In an embodiment, the candidate entity extraction module may be configured to perform the operation S520 described above, which is not described herein.

The dimension-reduction vector sequence obtaining unit may be configured to perform dimension reduction on the candidate entity by using the full connection layer to obtain a dimension-reduction vector sequence. In an embodiment, the dimension-reduced vector sequence obtaining unit may be configured to perform the operation S530 described above, which is not described herein.

The predictive tag sequence obtaining unit may be configured to extract sentence level information in the dimension-reduced vector sequence from the bidirectional long short-time memory layer, and output a predictive tag sequence. In an embodiment, the predicted tag sequence obtaining unit may be configured to perform the operation S540 described above, which is not described herein.

The first tag sequence obtaining unit may be configured to output the first tag sequence by using the rule constraint that the conditional random field layer performs knowledge point recognition on the predicted tag sequence. In an embodiment, the first tag sequence obtaining unit may be configured to perform the operation S550 described above, which is not described herein.

According to an embodiment of the invention, the candidate entity extraction module comprises an enumeration unit and a convolution unit.

The enumeration unit may be configured to enumerate all candidate entities using a convolution kernel based on the sequence of embedded vectors. In an embodiment, the enumeration unit may be configured to perform the operation S610 described above, which is not described herein.

The convolution unit may be configured to encode the candidate entity to the fully-connected layer by a convolution operation. In an embodiment, the convolution unit may be configured to perform the operation S620 described above, which is not described herein.

According to the embodiment of the invention, the relation extraction module can comprise a knowledge point relation extraction feature acquisition unit and a relation acquisition unit.

The knowledge point relation extraction feature obtaining unit may be configured to use the first tag sequence, the second tag sequence, and the third tag sequence as input, and perform relation extraction by using the transform model, to obtain knowledge point relation extraction features. In an embodiment, the knowledge point relation extraction feature obtaining unit may be configured to perform the operation S710 described above, which is not described herein.

The relationship obtaining unit may be configured to input the knowledge point relationship extraction feature to an activation function layer, and obtain a relationship between the knowledge points. In an embodiment, the relationship obtaining unit may be configured to perform the operation S720 described above, which is not described herein.

According to an embodiment of the present invention, any of the target course requirement obtaining module 910, the data processing module 920, the course content similarity obtaining module 930, and the course recommending module 940 may be combined into one module to be implemented, or any of the modules may be split into a plurality of modules. Or at least some of the functionality of one or more of the modules may be combined with, and implemented in, at least some of the functionality of other modules. According to embodiments of the present disclosure, at least one of the target course demand acquisition module 910, the data processing module 920, the course content similarity acquisition module 930, and the course recommendation module 940 may be implemented at least in part as hardware circuitry, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system on a chip, a system on a substrate, a system on a package, an Application Specific Integrated Circuit (ASIC), or in hardware or firmware in any other reasonable manner of integrating or packaging the circuitry, or in any one of or a suitable combination of three of software, hardware, and firmware. Or at least one of the target course requirement retrieval module 910, the data processing module 920, the course content similarity retrieval module 930, and the course recommendation module 940 may be implemented at least in part as computer program modules that, when executed, perform the corresponding functions.

As shown in fig. 10, an electronic device 1000 according to an embodiment of the present invention includes a processor 1001 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 1002 or a program loaded from a storage section 1008 into a Random Access Memory (RAM) 1003. The processor 1001 may include, for example, a general purpose microprocessor (e.g., a CPU), an instruction set processor and/or an associated chipset and/or a special purpose microprocessor (e.g., an Application Specific Integrated Circuit (ASIC)), or the like. The processor 1001 may also include on-board memory for caching purposes. The processor 1001 may include a single processing unit or a plurality of processing units for performing different actions of the method flow according to an embodiment of the invention.

In the RAM 1003, various programs and data necessary for the operation of the electronic apparatus 1000 are stored. The processor 1001, the ROM 1002, and the RAM 1003 are connected to each other by a bus 1004. The processor 1001 performs various operations of the method flow according to the embodiment of the present invention by executing programs in the ROM 1002 and/or the RAM 1003. Note that the program may be stored in one or more memories other than the ROM 1002 and the RAM 1003. The processor 1001 may also perform various operations of the method flow according to an embodiment of the present invention by executing programs stored in the one or more memories.

According to an embodiment of the invention, the electronic device 1000 may further comprise an input/output (I/O) interface 1005, the input/output (I/O) interface 1005 also being connected to the bus 1004. The electronic device 1000 may also include one or more of the following components connected to the I/O interface 1005: an input section 1006 including a keyboard, a mouse, and the like; an output portion 1007 including a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), etc., and a speaker, etc.; a storage portion 1008 including a hard disk or the like; and a communication section 1009 including a network interface card such as a LAN card, a modem, or the like. The communication section 1009 performs communication processing via a network such as the internet. The drive 1010 is also connected to the I/O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like, is installed as needed in the drive 1010, so that a computer program read out therefrom is installed as needed in the storage section 1008.

The present invention also provides a computer-readable storage medium that may be embodied in the apparatus/device/system described in the above embodiments; or may exist alone without being assembled into the apparatus/device/system. The computer-readable storage medium carries one or more programs which, when executed, implement methods in accordance with embodiments of the present invention.

According to embodiments of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, which may include, for example, but is not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to embodiments of the invention, the computer-readable storage medium may include ROM 1002 and/or RAM1003 described above and/or one or more memories other than ROM 1002 and RAM 1003.

Embodiments of the present invention also include a computer program product comprising a computer program containing program code for performing the method shown in the flowcharts. The program code means for causing a computer system to carry out the methods provided by embodiments of the present invention when the computer program product is run on the computer system.

The above-described functions defined in the system/apparatus of the embodiment of the present invention are performed when the computer program is executed by the processor 1001. The systems, apparatus, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the invention.

In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, or the like. In another embodiment, the computer program may also be transmitted in the form of signals on a network medium, distributed, and downloaded and installed via the communication section 1009, and/or installed from the removable medium 1011. The computer program may include program code that may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

In such an embodiment, the computer program may be downloaded and installed from a network via the communication portion 1009, and/or installed from the removable medium 1011. The above-described functions defined in the system of the embodiment of the present invention are performed when the computer program is executed by the processor 1001. The systems, devices, apparatus, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the invention.

According to embodiments of the present invention, program code for carrying out computer programs provided by embodiments of the present invention may be written in any combination of one or more programming languages, and in particular, such computer programs may be implemented in high-level procedural and/or object-oriented programming languages, and/or in assembly/machine languages. Programming languages include, but are not limited to, such as Java, c++, python, "C" or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The embodiments of the present invention are described above. These examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Although the embodiments are described above separately, this does not mean that the measures in the embodiments cannot be used advantageously in combination. Various alternatives and modifications can be made by those skilled in the art without departing from the scope of the invention, and such alternatives and modifications are intended to fall within the scope of the invention.

Claims

1. A course recommendation method, the method comprising:

acquiring a target course requirement of a target user;

performing data processing on the target course demands to obtain target course data;

Traversing the target course data by using a course content knowledge graph to obtain course content similarity, wherein the course content knowledge graph comprises a plurality of entities, a plurality of edges and importance weights corresponding to the entities, the entities are knowledge points obtained by identifying multi-mode platform course information by using a knowledge point identification model and a knowledge point corresponding model, and the edges are relations among the knowledge points; and

And recommending courses with the similarity of the course content larger than a preset threshold value to the target user.

2. The method of claim 1, wherein the multimodal platform lesson information comprises text information, image information, and audio information, and wherein the recognition of the plurality of entities based on the multimodal platform lesson information using a knowledge point recognition model and a knowledge point correspondence model specifically comprises:

Performing first preprocessing on the text information to obtain sequence labeling data divided in sentence level;

inputting the sequence labeling data into the knowledge point recognition model, and outputting a first label sequence corresponding to the text information;

Performing second preprocessing on the image information to obtain image characteristics;

performing third preprocessing on the audio information to obtain audio characteristics;

Inputting the image features and the audio features into the knowledge point corresponding model to perform knowledge point mapping to obtain a second tag sequence and a third tag sequence; and

The plurality of entities is obtained based on the first tag sequence, the second tag sequence, and the third tag sequence.

3. The method according to claim 2, wherein the knowledge point recognition model includes an embedded layer, a convolutional neural network layer, a full-connection layer, a two-way long and short-term memory layer and a conditional random field layer which are connected in series, the sequence labeling data is input into the knowledge point recognition model, and a first tag sequence corresponding to text information is output, and the method specifically includes:

The embedded layer converts the sequence labeling data into an embedded vector sequence;

The convolutional neural network layer extracts candidate entities of the embedded vector sequence;

The full connection layer performs dimension reduction on the candidate entity to obtain a dimension reduction vector sequence;

the two-way long short-time memory layer extracts sentence level information in the dimension reduction vector sequence and outputs a prediction tag sequence; and

And the conditional random field layer carries out rule constraint of knowledge point identification on the predicted tag sequence and outputs the first tag sequence.

4. A method according to claim 3, wherein the convolutional neural network layer sets a convolutional kernel having a plurality of convolutional kernel sizes, and wherein the convolutional neural network layer extracts candidate entities of the embedded vector sequence, comprising:

enumerating all candidate entities using the convolution kernel based on the embedded vector sequence; and

The candidate entity is encoded to the fully-connected layer by a convolution operation.

5. The method according to any one of claims 2 to 4, wherein the relationship between knowledge points is obtained by using a transducer model, and wherein the method specifically comprises:

Taking the first tag sequence, the second tag sequence and the third tag sequence as input, and performing relation extraction by using the transducer model to obtain knowledge point relation extraction features; and

And inputting the knowledge point relation extraction features into an activation function layer to acquire the relation among the knowledge points.

6. The method of claim 1, wherein the importance weights are determined based on the frequency of occurrence of knowledge points.

7. The method according to claim 1, wherein the data processing is performed on the target course requirement to obtain target course data, and the method specifically includes:

Extracting audio text of the target course requirement and cleaning the text under the condition that the target course requirement is in an audio form;

Performing text cleaning on the target course requirement under the condition that the target course requirement is in a text form;

extracting keywords and phrases from the target course requirement after text cleaning to obtain key information; and

And carrying out format conversion on the key information to obtain the target course data.

8. A course recommendation device, the device comprising:

the target course demand acquisition module is used for: acquiring a target course requirement of a target user;

A data processing module for: performing data processing on the target course demands to obtain target course data;

The course content similarity acquisition module is used for: traversing the target course data by using a course content knowledge graph to obtain course content similarity, wherein the course content knowledge graph comprises a plurality of entities, a plurality of edges and importance weights corresponding to the entities, the entities are knowledge points obtained by identifying multi-mode platform course information by using a knowledge point identification model and a knowledge point corresponding model, and the edges are relations among the knowledge points; and

Course recommendation module for: and recommending courses with the similarity of the course content larger than a preset threshold value to the target user.

9. An electronic device, comprising:

one or more processors;

Storage means for storing one or more programs,

Wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the method of any of claims 1-7.

10. A computer readable storage medium having stored thereon executable instructions which, when executed by a processor, cause the processor to perform the method according to any of claims 1-7.