Learning content recommendation method and system based on artificial intelligence

Abstract

The present invention provides a method and system for recommending learning content based on artificial intelligence, which relates to the field of artificial intelligence technology. In the present invention, first, a learning content distribution map is constructed; second, based on two learning contents of interest, the content of interest is expanded in the learning content distribution map to form at least one extended combination of content of interest; then, using the target learning content recommendation network, each extended combination of content of interest is respectively subjected to content recommendation processing to obtain corresponding content recommendation results; finally, based on the corresponding content recommendation results, a target extended combination of content of interest is determined in at least one extended combination of content of interest, and based on the learning content included in the target extended combination of content of interest, a learning content recommendation operation is performed on the target learning user. Based on the above method, the problem of low reliability of learning content recommendation existing in the prior art can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based learning content recommendation method and system. Background Art

[0002] In many learning application scenarios such as language learning, it is necessary to recommend learning content to users (such as students) so that users can learn effectively. For example, in English learning, it is necessary to learn learning content such as words, sentences and articles. Therefore, it is necessary to recommend corresponding learning content such as words, sentences and articles to users according to their actual needs. However, in the prior art, learning content is generally arbitrarily determined from content that the user has not learned, and recommended to the user for learning. In this way, there is a problem of low reliability of learning content recommendation. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a learning content recommendation method and system based on artificial intelligence to improve the problem of low reliability of learning content recommendation in the prior art.

[0004] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0005] A learning content recommendation method based on artificial intelligence, comprising:

[0006] Based on the multiple candidate learning contents, a corresponding learning content distribution map is constructed, wherein the learning content distribution map is composed of the multiple candidate learning contents and the association relationships between the multiple candidate learning contents, and in the learning content distribution map, the name of the map node is the name of the corresponding candidate learning content, and the attribute information of the map node is the corresponding candidate learning content;

[0007] Determine two candidate learning contents from the plurality of candidate learning contents as the learning contents of interest to the target learning user, and based on the two learning contents of interest, perform content of interest expansion in the learning content distribution map to form at least one content of interest expansion combination, wherein each of the content of interest expansion combinations includes the two learning contents of interest;

[0008] Using the target learning content recommendation network, each of the interest content extension combinations is subjected to content recommendation processing to obtain corresponding content recommendation results, wherein the target learning content recommendation network is a neural network formed through network optimization, and the content recommendation results are used to reflect the recommendation degree of the learning content included in the corresponding interest content extension combination;

[0009] Based on the corresponding content recommendation result, a target content extension combination of interest is determined in the at least one content extension combination of interest, and based on the learning content included in the target content extension combination of interest, a learning content recommendation operation is performed on the target learning user.

[0010] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning content recommendation method, the step of constructing a corresponding learning content distribution map based on multiple candidate learning contents includes:

[0011] Based on the multiple candidate learning contents, determining a corresponding multiple graph nodes, wherein there is a one-to-one correspondence between the multiple graph nodes and the multiple candidate learning contents;

[0012] For every two graph nodes, the semantic similarity of the two candidate learning contents corresponding to the two graph nodes is calculated to obtain the semantic similarity corresponding to the two graph nodes;

[0013] For every two graph nodes, the overlap degree of the learning user sets corresponding to the two graph nodes is calculated to obtain the user overlap degree corresponding to the two graph nodes, wherein the learning user set refers to the set of historical learning users who have studied or recommended in history;

[0014] For every two graph nodes, based on the semantic similarity and user overlap corresponding to the two graph nodes, determine whether there is a correlation between the two graph nodes, and when there is no correlation between the two graph nodes, determine not to associate the two graph nodes. Alternatively, when there is a correlation between the two graph nodes, associate the two graph nodes to form a corresponding learning content distribution map, wherein in the learning content distribution map, there is an associated line segment between the two associated graph nodes.

[0015] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning content recommendation method, the step of determining two candidate learning contents from the multiple candidate learning contents as the learning contents of interest to the target learning user, and expanding the contents of interest in the learning content distribution map based on the two learning contents of interest to form at least one expanded combination of contents of interest includes:

[0016] In response to a selection operation performed by a target learning user, two candidate learning contents are determined from the plurality of candidate learning contents as learning contents of interest to the target learning user, or, based on historical learning data of the target learning user, two candidate learning contents are determined from the plurality of candidate learning contents as learning contents of interest to the target learning user;

[0017] Determine one of the two learning contents of interest as a first learning content of interest, and determine the other learning content of interest of the two learning contents of interest as a second learning content of interest;

[0018] Starting from the first learning content of interest, wandering in the learning content distribution map until wandering to the second learning content of interest, so as to form at least one wandering sequence, wherein in each wandering sequence, no two identical candidate learning contents or map nodes are included, and when multiple wandering sequences are formed, every two wandering sequences are at least partially different;

[0019] At least one content-of-interest extension combination is formed based on the at least one walking sequence.

[0020] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning content recommendation method, the step of using the target learning content recommendation network to perform content recommendation processing on each of the interest content extension combinations to obtain corresponding content recommendation results includes:

[0021] Using a target learning content recommendation network, performing a content mining operation on the at least one interest content extension combination to form a learning content global vector set and a learning content extension vector for each interest content extension combination in the at least one interest content extension combination, wherein the learning content global vector in the learning content global vector set serves as a semantic vector of each candidate learning content in the corresponding interest content extension combination, and the learning content extension vector serves as a semantic vector of an associated extended learning content of each candidate learning content in the corresponding interest content extension combination in the learning content distribution map;

[0022] For each of the at least one interest content extension combination, using the target learning content recommendation network, a learning content global vector set and a learning content extension vector corresponding to the interest content extension combination are focused and merged to form a learning content focused merged vector corresponding to the interest content extension combination;

[0023] For each of the at least one interest content extension combination, the target learning content recommendation network is used to perform content recommendation processing on the learning content focus merging vector corresponding to the interest content extension combination to form a content recommendation result corresponding to the interest content extension combination.

[0024] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning content recommendation method, the step of using the target learning content recommendation network to perform content mining operations on the at least one interesting content extension combination to form a learning content global vector set and a learning content extension vector for each interesting content extension combination in the at least one interesting content extension combination includes:

[0025] For any of the interest content extension combinations, the following operations are performed:

[0026] Using the content mining subnetwork included in the target learning content recommendation network to perform content mining operations, forming candidate learning content semantic vectors and candidate learning content distribution vectors corresponding to each candidate learning content in the interest content extension combination, and forming a learning content extension vector corresponding to the interest content extension combination, wherein the candidate learning content semantic vector is used to reflect the semantic information of the candidate learning content itself, and the candidate learning content distribution vector is used to reflect the distribution relationship information of the candidate learning content in the interest content extension combination;

[0027] By utilizing the content association merging subnetwork included in the target learning content recommendation network, each candidate learning content semantic vector, each candidate learning content distribution vector and the learning content extension vector are associated and merged to form a global learning content vector set corresponding to the interested content extension combination.

[0028] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning content recommendation method, the step of performing content mining operations using the content mining subnetwork included in the target learning content recommendation network to form candidate learning content semantic vectors and candidate learning content distribution vectors corresponding to each candidate learning content in the interest content extension combination, and forming the learning content extension vector corresponding to the interest content extension combination, includes:

[0029] Using the content mining unit in the content mining subnetwork included in the target learning content recommendation network, each candidate learning content in the extended combination of interested content and the distribution relationship information of each candidate learning content are mapped into a vector space to form a candidate learning content semantic vector corresponding to each candidate learning content and a candidate learning content distribution vector corresponding to each candidate learning content;

[0030] Utilizing the extended combination mining unit in the content mining subnetwork, the associated extended learning contents of the extended content combination of interest in the learning content distribution map are mapped into a vector space to form a learning content extension vector corresponding to the extended content combination of interest, wherein the associated extended learning contents include other candidate learning contents of each associated configuration of the candidate learning contents in the extended content combination of interest in the learning content distribution map.

[0031] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning content recommendation method, the step of using the content association merging subnetwork included in the target learning content recommendation network to associate and merge each candidate learning content semantic vector, each candidate learning content distribution vector and the learning content extension vector to form a global learning content vector set corresponding to the interested content extension combination includes:

[0032] Utilizing the content association merging subnetwork included in the target learning content recommendation network, respectively merging each candidate learning content semantic vector and the corresponding candidate learning content distribution vector, and outputting the corresponding content semantic distribution merging vector;

[0033] Based on the corresponding learning content extension vectors, focus merging operations are performed on each of the content semantic distribution merge vectors to form corresponding learning content global vectors, and based on each of the learning content global vectors, a learning content global vector set corresponding to the interested content extension combination is determined.

[0034] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning content recommendation method, the step of performing a focus merging operation on the learning content global vector set and the learning content extension vector corresponding to the interested content extension combination using the target learning content recommendation network for each interested content extension combination in the at least one interested content extension combination to form a learning content focus merging vector corresponding to the interested content extension combination includes:

[0035] For each of the at least one extended combination of content of interest, use the first spatial mapping parameter, the second spatial mapping parameter and the third spatial mapping parameter included in the target learning content recommendation network to spatially map each of the global learning content vectors and the learning content extended vector in the global learning content vector set corresponding to the extended combination of content of interest, output the first spatial mapping result, the second spatial mapping result and the third spatial mapping result corresponding to each of the global learning content vectors, and, based on the first spatial mapping result, the second spatial mapping result and the third spatial mapping result, merge and output the extended combination association parameter corresponding to the global learning content vector, so as to obtain the extended combination association parameter corresponding to each of the global learning content vectors in the extended combination of content of interest;

[0036] Based on the extended combination relevance parameter corresponding to each of the learning content global vectors in an extended combination of the content of interest and each of the learning content global vectors, a learning content focus merged vector of the extended combination of the content of interest is merged and output.

[0037] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning content recommendation method, the step of performing content recommendation processing on the learning content focus merging vector corresponding to the at least one interesting content extension combination using the target learning content recommendation network to form a content recommendation result corresponding to the interesting content extension combination for each interesting content extension combination in the at least one interesting content extension combination includes:

[0038] Extracting a local learning content distribution map corresponding to the at least one extended combination of content of interest from the learning content distribution map, wherein the local learning content distribution map includes each candidate learning content in the at least one extended combination of content of interest and a corresponding association relationship;

[0039] Performing content mining operations on the local learning content distribution map, and outputting a corresponding local map semantic vector, wherein the local map semantic vector is used to reflect the semantic information of each candidate learning content in the local learning content distribution map and the semantic information of the corresponding association relationship;

[0040] Based on the number of the at least one combination of content extensions of interest, a corresponding number of random noise vectors are generated, wherein the size of each of the random noise vectors is consistent with the size of the local graph semantic vector, and for any one of the random noise vectors, the number of non-zero parameters in the random noise vector is less than the number of zero parameters, and the number of non-zero parameters between any two of the random noise vectors is the same, the distribution shape is the same, and the distribution positions are not completely the same;

[0041] Superimposing each of the random noise vectors and the local graph semantic vector respectively to form at least one corresponding local graph noise vector, and assigning the at least one local graph noise vector to the at least one content of interest extension combination, so that a one-to-one correspondence is formed between the at least one local graph noise vector and the at least one content of interest extension combination;

[0042] For each of the at least one content extension combination of interest, based on the local graph noise vector corresponding to the content extension combination of interest, a focusing operation is performed on the learning content focus merging vector corresponding to the content extension combination of interest to form a corresponding target vector, and the target learning content recommendation network is used to perform content recommendation processing on the target vector to form a content recommendation result corresponding to the content extension combination of interest.

[0043] An embodiment of the present invention also provides an artificial intelligence-based learning content recommendation system, including a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-mentioned artificial intelligence-based learning content recommendation method.

[0044] In the artificial intelligence-based learning content recommendation method and system provided in the embodiments of the present invention, first, a learning content distribution map is constructed; secondly, based on two interested learning contents, the interested contents are expanded in the learning content distribution map to form at least one interested content expansion combination; then, using the target learning content recommendation network, content recommendation processing is performed on each interested content expansion combination respectively to obtain corresponding content recommendation results; finally, based on the corresponding content recommendation results, a target interested content expansion combination is determined in at least one interested content expansion combination, and based on the learning content included in the target interested content expansion combination, a learning content recommendation operation is performed on the target learning user. Based on the above method, since a learning content distribution map will be constructed, the reliability of the expansion of the content of interest based on the learning content distribution map is higher (that is, the association relationship in the learning content distribution map is fully utilized for expansion, so that compared with the conventional arbitrary expansion, it can have higher reliability), and, after obtaining at least one content of interest, the recommendation degree will be evaluated based on the target learning content recommendation network, so that the target content of interest expansion combination can be determined based on the obtained content recommendation results. In this way, the reliability of the learning content recommended to the target learning user can be further improved (compared with the conventional arbitrary selection of learning content, it can have higher reliability), thereby improving the problem of low reliability of learning content recommendation in the prior art, so that learning resources can be effectively utilized and the problem of resource waste can be avoided. Therefore, it also has a high application value in practical applications.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A structural block diagram of an artificial intelligence-based learning content recommendation system provided in an embodiment of the present invention.

[0047] Figure 2 A flowchart of the steps of the artificial intelligence-based learning content recommendation method provided in an embodiment of the present invention.

[0048] Figure 3 A schematic diagram of a learning content distribution map provided in an embodiment of the present invention.

[0049] Figure 4 A schematic diagram of a focus merging operation provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, an embodiment of the present invention provides a learning content recommendation system based on artificial intelligence. Among them, the learning content recommendation system based on artificial intelligence may include a memory and a processor. In detail, the memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, they can be electrically connected to each other through one or more communication buses or signal lines. The memory may store at least one software function module (computer program) that can exist in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby realizing the learning content recommendation method based on artificial intelligence provided by the embodiment of the present invention (as described later).

[0053] Optionally, the memory may be, but not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. The processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0054] and, Figure 1 The structure shown is for illustration only. The artificial intelligence-based learning content recommendation system may also include Figure 1 More or fewer components as shown, or with Figure 1 The various configurations shown may, for example, include a communication unit for exchanging information with other devices.

[0055] In an alternative example, the artificial intelligence-based learning content recommendation system may be a server with data processing capabilities.

[0056] Combination Figure 2 The embodiment of the present invention also provides an artificial intelligence-based learning content recommendation method, which can be applied to the artificial intelligence-based learning content recommendation system. The method steps defined in the process related to the artificial intelligence-based learning content recommendation method can be implemented by the artificial intelligence-based learning content recommendation system.

[0057] The following will Figure 2 The specific process shown is explained in detail.

[0058] Step S110, constructing a corresponding learning content distribution map based on multiple candidate learning contents.

[0059] In an embodiment of the present invention, the artificial intelligence-based learning content recommendation system can construct a corresponding learning content distribution map based on multiple candidate learning contents. Wherein, the learning content distribution map is composed of the multiple candidate learning contents and the association relationship between the multiple candidate learning contents, and in the learning content distribution map, the name of the map node is the name of the corresponding candidate learning content, and the attribute information of the map node is the corresponding candidate learning content. Exemplarily, the candidate learning content may refer to English words, English sentences or English articles, etc. For example, in the application scenario where the user learns words, the multiple candidate learning contents may refer to multiple English words, in the scenario where the user learns sentences, the multiple candidate learning contents may refer to multiple English sentences, and in the scenario where the user learns articles, the multiple candidate learning contents may refer to multiple English articles. In addition, when English words are used as candidate learning contents, the English words themselves can be used as names, and the Chinese interpretations, etc. can be used as corresponding specific learning contents, that is, as the attribute information of the corresponding map nodes. When an English sentence is used as a candidate learning content, the key words in it can be used as the name, and the English sentence itself and the corresponding Chinese content can be used as the corresponding specific learning content, that is, as the attribute information of the corresponding graph node. When an English article is used as a candidate learning content, the English title can be used as the name, and the English article itself and the corresponding Chinese content can be used as the corresponding specific learning content, that is, as the attribute information of the corresponding graph node.

[0060] For example, for Abandon, the corresponding candidate learning content may include:

[0061] Abandon is a polysemous word, mainly used in the following different contexts:

[0062] Verb: Abandon: to give up completely a plan, idea, or responsibility. For example, “She decided to abandon her original project due to unexpected circumstances.” Abandon: to abandon something or someone, usually because it is no longer possible to care for or afford it. For example, “The family had to abandon their home due to the flood.” Stop: to stop supporting or paying attention to something. For example, “The company abandoned the new software development after it failed initial testing.”

[0063] Noun: Indulgence: refers to a behavior or attitude that is completely free or unrestrained. For example, "The party was fil led with abandon, with everyone dancing and enjoying themselves."

[0064] Meaning details:Emotions and attitudes: As a noun, "abandon" can indicate a state of enthusiasm, excitement, or recklessness, such as giving full play to music or dance;Law and responsibilities: In a legal context, "abandon" may involve giving up a right, property, or obligation;

[0065] Usage examples: "After a long debate, they chose to abandon the project and start a new one." "The sense of abandon was palpable in the crowd during the concert."

[0066] Step S120, determining two candidate learning contents from the multiple candidate learning contents as the interested learning contents of the target learning user, and based on the two interested learning contents, performing interested content expansion in the learning content distribution map to form at least one interested content expansion combination.

[0067] In an embodiment of the present invention, the artificial intelligence-based learning content recommendation system can determine two candidate learning contents from the multiple candidate learning contents as the learning contents of interest to the target learning user, and based on the two learning contents of interest, expand the content of interest in the learning content distribution map to form at least one content of interest expansion combination. Each of the content of interest expansion combinations includes two learning contents of interest, that is, the content of interest expansion combination includes at least two learning contents of interest, that is, on the basis of including the two learning contents of interest, it may also include other candidate learning contents expanded based on the learning content distribution map, so that the learning content is richer. For example, for the above-mentioned Abandon, the names of other candidate learning contents with associated relationships that are expanded may include:

[0068] Forsake: means to give up or abandon, usually with an emotional distance;

[0069] Desert: usually refers to leaving or abandoning someone or something when needed;

[0070] Re li nqu i sh: means to voluntarily give up certain rights or control;

[0071] Leave: refers to leaving a place or thing, and is sometimes used to mean giving up;

[0072] Qu it: means to give up an activity or responsibility, usually meaning to terminate;

[0073] Renounce: To formally renounce a right, belief, or responsibility.

[0074] Step S130 , using the target learning content recommendation network, performing content recommendation processing on each of the interest content extension combinations to obtain corresponding content recommendation results.

[0075] In an embodiment of the present invention, the artificial intelligence-based learning content recommendation system can use the target learning content recommendation network to perform content recommendation processing on each of the interest content extension combinations, and obtain corresponding content recommendation results. Wherein, the target learning content recommendation network is a neural network formed by network optimization (for example, the network parameters can be updated and optimized based on the training learning content data and the corresponding actual recommendation degree (i.e., label)), and the content recommendation result is used to reflect the recommendation degree of the learning content included in the corresponding interest content extension combination. Exemplarily, the content recommendation result can have different forms according to the different output layer functions of the target learning content recommendation network. For example, each recommendation score of 0-100 can be output through a regression function, or a variety of classification results can be output through classification functions such as softmax, such as extremely high recommendation degree, high recommendation degree, general recommendation degree, low recommendation degree, extremely low recommendation degree, etc.

[0076] Step S140, based on the corresponding content recommendation result, determine a target interest content extension combination in the at least one interest content extension combination, and perform a learning content recommendation operation on the target learning user based on the learning content included in the target interest content extension combination.

[0077] In an embodiment of the present invention, the artificial intelligence-based learning content recommendation system can determine a target interest content extension combination in the at least one interest content extension combination based on the corresponding content recommendation result, and perform a learning content recommendation operation on the target learning user based on the learning content included in the target interest content extension combination. Exemplarily, the learning content included in the interest content extension combination with the highest recommendation degree represented by the corresponding content recommendation result in the at least one interest content extension combination can be pushed to the target learning user, so that the target learning user can learn the learning content. In this way, it is ensured that the target learning user can learn the learning content of interest and can also learn more related learning content.

[0078] Based on the above method, since a learning content distribution map will be constructed, the reliability of the expansion of the content of interest based on the learning content distribution map is higher (that is, the association relationship in the learning content distribution map is fully utilized for expansion, so that compared with the conventional arbitrary expansion, it can have higher reliability), and, after obtaining at least one content of interest, the recommendation degree will be evaluated based on the target learning content recommendation network, so that the target content of interest expansion combination can be determined based on the obtained content recommendation results. In this way, the reliability of the learning content recommended to the target learning user can be further improved (compared with the conventional arbitrary selection of learning content, it can have higher reliability), thereby improving the problem of low reliability of learning content recommendation in the prior art, so that learning resources can be effectively utilized and the problem of resource waste can be avoided. Therefore, it also has a high application value in practical applications.

[0079] In the above step S110, the specific method of constructing the corresponding learning content distribution map is not limited and can be selected according to actual needs.

[0080] For example, in a specific implementation, in order to ensure that the association relationship between the graph nodes in the constructed learning content distribution graph is more reliable, that is, to ensure that the association relationship between the candidate learning contents is more reliable, so that the reliability of subsequent association expansion is higher, the above step S110 may further include the following detailed process:

[0081] First, based on the multiple candidate learning contents, a corresponding multiple graph nodes can be determined, wherein there is a one-to-one correspondence between the multiple graph nodes and the multiple candidate learning contents, that is, one candidate learning content can be used as a graph node;

[0082] Secondly, for every two graph nodes, the semantic similarity of the two candidate learning contents corresponding to the two graph nodes can be calculated to obtain the semantic similarity corresponding to the two graph nodes. For example, the edit distance of the two candidate learning contents can be calculated, and the semantic similarity can be determined based on the edit distance, wherein the smaller the edit distance, the greater the corresponding semantic similarity;

[0083] Then, for each two graph nodes, the overlap degree of the learning user sets corresponding to the two graph nodes is calculated to obtain the user overlap degree corresponding to the two graph nodes, wherein the learning user set refers to the set of historical learning users who have studied or recommended in history, or, in other implementations, the learning user set may also refer to the set of historical learning users who have historically used the corresponding candidate learning content as interested learning content;

[0084] Finally, for every two graph nodes, based on the semantic similarity and user overlap corresponding to the two graph nodes, determine whether the two graph nodes have a correlation relationship, and when the two graph nodes do not have a correlation relationship, determine not to associate the two graph nodes, or, when the two graph nodes have a correlation relationship, associate the two graph nodes to form a corresponding learning content distribution map; illustratively, the semantic similarity and the user overlap can be averaged or weighted averaged to obtain the corresponding correlation degree, and then, two graph nodes with a correlation degree greater than a preset correlation degree (such as 0.5, 0.6, 0.7, etc.) are determined to have a correlation relationship, and two graph nodes with a correlation degree not greater than the preset correlation degree are determined to have no correlation relationship, wherein, in the learning content distribution map, there is an associated line segment between the two graph nodes in the associated configuration, such as Figure 3 shown.

[0085] In the above step S120, the specific method of forming at least one extended combination of content of interest is not limited and can be selected according to actual needs.

[0086] For example, in a specific implementation, in order to make the at least one extended combination of interested contents have high reliability and rich content to ensure the reliability of subsequent learning content recommendation, the above step S120 may further include the following detailed process:

[0087] First, in response to the selection operation performed by the target learning user, two candidate learning contents can be determined from the multiple candidate learning contents as the learning contents of interest to the target learning user, or, based on the historical learning data of the target learning user, two candidate learning contents can be determined from the multiple candidate learning contents as the learning contents of interest to the target learning user. That is to say, in one application scenario, the corresponding learning contents of interest can be determined based on the selection of the target learning user, such as the content that the target learning user inputs to learn; in another application scenario, it is necessary to make corresponding predictions in combination with the historical learning data of the target learning user. For example, the content that has been learned with high frequency in history can be determined based on the historical learning data;

[0088] Secondly, one of the two learning contents of interest may be determined as a first learning content of interest, and the other of the two learning contents of interest may be determined as a second learning content of interest;

[0089] Then, starting from the first learning content of interest, wandering in the learning content distribution map until wandering to the second learning content of interest, so as to form at least one wandering sequence, wherein in each wandering sequence, two identical candidate learning contents or map nodes are not included, and when multiple wandering sequences are formed, each two wandering sequences are at least partially different, that is, except for the first learning content of interest and the second learning content of interest, other learning contents may be completely different or partially different;

[0090] Finally, at least one interesting content extension combination may be formed based on the at least one walking sequence. For example, one walking sequence may be used as an interesting content extension combination.

[0091] Among them, it can be understood that in the above steps, when walking, all the walking sequences can be determined, and part of the walking sequences can also be determined. For example, in order to avoid the problem of excessive resource consumption caused by too many walking sequences, part of the walking sequences can be determined, that is, the walking process is specifically limited. For example, the walking can be based on the following method: during the walking process, the correlation between the two adjacent learning contents walked to is greater than the configured correlation threshold, and during the walking process, the number of learning contents between the first learning content of interest and the learning content currently walked to is not greater than the configured number threshold.

[0092] In the above step S130, the specific manner of performing content recommendation processing on each of the extended combinations of content of interest is not limited and can be selected according to actual needs.

[0093] For example, in a specific implementation, in order to improve the reliability of content recommendation processing, the above-mentioned step S130 may include step S131, step S132 and step S133.

[0094] Step S131, using the target learning content recommendation network, performs content mining operations on the at least one interested content extension combination to form a learning content global vector set and a learning content extension vector for each interested content extension combination in the at least one interested content extension combination, wherein the learning content global vector in the learning content global vector set serves as the semantic vector of each candidate learning content in the corresponding interested content extension combination.

[0095] In an embodiment of the present invention, a target learning content recommendation network can be used to perform a content mining operation on the at least one interested content extension combination to form a learning content global vector set and a learning content extension vector for each interested content extension combination in the at least one interested content extension combination, wherein the learning content global vector in the learning content global vector set serves as a semantic vector of each candidate learning content in the corresponding interested content extension combination (that is, a content mining operation can be performed on the candidate learning content to obtain a corresponding semantic vector), and the learning content extension vector serves as a semantic vector of an associated extended learning content of each candidate learning content in the corresponding interested content extension combination in the learning content distribution map (that is, a content mining operation can be performed on the associated extended learning content to obtain a corresponding semantic vector).

[0096] Step S132: For each of the at least one interested content extension combination, using the target learning content recommendation network, a focus merging operation is performed on the learning content global vector set and the learning content extension vector corresponding to the interested content extension combination to form a learning content focus merging vector corresponding to the interested content extension combination.

[0097] In an embodiment of the present invention, for each of the at least one interest content extension combination, the target learning content recommendation network is used to perform a focus merging operation on the learning content global vector set and the learning content extension vector corresponding to the interest content extension combination to form a learning content focus merging vector corresponding to the interest content extension combination. In this way, the learning content global vector set and the learning content extension vector can be fully and reliably merged, thereby ensuring that the obtained learning content focus merging vector has a better representation capability for the learning content of the corresponding interest content extension combination.

[0098] Step S133, for each of the at least one interested content extension combination, using the target learning content recommendation network, perform content recommendation processing on the learning content focus merging vector corresponding to the interested content extension combination to form a content recommendation result corresponding to the interested content extension combination.

[0099] In an embodiment of the present invention, for each of the at least one interest content extension combination, the target learning content recommendation network is used to perform content recommendation processing on the learning content focus merging vector corresponding to the interest content extension combination to form a content recommendation result corresponding to the interest content extension combination. Exemplarily, the fully connected subnetwork in the target learning content recommendation network can be used to perform full connection processing on the learning content focus merging vector corresponding to the interest content extension combination, and then the classification function (such as softmax function, etc.) or regression function in the target learning content recommendation network is used to classify or regress the vector output by the full connection processing to obtain the content recommendation result corresponding to the interest content extension combination.

[0100] It can be understood that, in the above-mentioned step S131, the specific manner of performing content mining operations on the at least one content extension combination of interest is not limited. For example, in a specific implementation, in order to ensure the accuracy of the content mining operation and make the information representation capability of the mined vector higher, the above-mentioned step S131 can further include step S131 a and step S131 b (for example, step S131 a and step S131 b can be executed for any one of the content extension combinations of interest), and the contents of each step are described as follows.

[0101] Step S131 a, using the content mining subnetwork included in the target learning content recommendation network to perform content mining operations, forming candidate learning content semantic vectors and candidate learning content distribution vectors corresponding to each candidate learning content in the interested content extension combination, and forming a learning content extension vector corresponding to the interested content extension combination.

[0102] In an embodiment of the present invention, after forming the extended combination of the content of interest, the content mining subnetwork included in the target learning content recommendation network can be used to perform content mining operations to form candidate learning content semantic vectors and candidate learning content distribution vectors corresponding to each candidate learning content in the extended combination of the content of interest, and to form a learning content extension vector corresponding to the extended combination of the content of interest. The candidate learning content semantic vector is used to reflect the semantic information of the candidate learning content itself, and the candidate learning content distribution vector is used to reflect the distribution relationship information of the candidate learning content in the extended combination of the content of interest (i.e., the position in the corresponding walk sequence, such as the first position, the second position, the third position, etc.).

[0103] Step S131b, using the content association merging subnetwork included in the target learning content recommendation network, associates and merges each candidate learning content semantic vector, each candidate learning content distribution vector and the learning content extension vector to form a global learning content vector set corresponding to the interested content extension combination.

[0104] In an embodiment of the present invention, after mining out candidate learning content semantic vectors, candidate learning content distribution vectors and learning content extension vectors, the content association merging subnetwork included in the target learning content recommendation network can be used to associate and merge each candidate learning content semantic vector, each candidate learning content distribution vector and the learning content extension vector to form a global learning content vector set corresponding to the content extension combination of interest. In this way, the global learning content vector set can be used to express information comprehensively and reliably.

[0105] It can be understood that, in the above step S131a, the specific manner of performing the content mining operation using the content mining subnetwork included in the target learning content recommendation network is not limited. For example, in a specific implementation, the above step S131a may include the following specific sub-steps that can be implemented:

[0106] First, using the content mining unit in the content mining subnetwork included in the target learning content recommendation network, each candidate learning content in the extended combination of interested content and the distribution relationship information of each candidate learning content are mapped into vector space, so as to form a candidate learning content semantic vector corresponding to each candidate learning content and a candidate learning content distribution vector corresponding to each candidate learning content; that is, the candidate learning content is mapped into vector space to obtain the candidate learning content semantic vector; the distribution relationship information is mapped into vector space to obtain the candidate learning content distribution vector; in addition, the vector space mapping may include word embedding processing and convolution processing. Specifically, after the candidate learning content is word embedded, the obtained word embedding vector may be convolution processed. For example, the candidate learning content may be "completely abandon a plan", and then, a pre-trained Word2Vec model is used to process "completely abandon a plan". The following is an example of the embedding process and results: word segmentation results: "thoroughly", "give up", "something", "plan"; specific content of the embedding vector: embedding vector of "thoroughly": [0.1,-0.3,0.5,...]; embedding vector of "give up": [0.2,0.4,-0.1,...]; embedding vector of "something": [0.3,0.2,0.6,...]; embedding vector of "plan": [0.8,0.1,-0.2,...]; then, these embedding vectors can be concatenated to obtain {[0.1,-0.3,0.5,...]; [0.2,0.4,-0.1,...]; [0.3,0.2,0.6,...]; [0.8,0.1,-0.2,...]}, then, the concatenated vectors can be convolved with the convolution kernel to obtain the candidate learning content semantic vector;

[0107] Secondly, the extended combination mining unit in the content mining subnetwork can be used to perform vector space mapping on the associated extended learning content of the extended combination of content of interest in the learning content distribution map to form a learning content extension vector corresponding to the extended combination of content of interest, wherein the associated extended learning content includes other candidate learning content of each associated configuration of each candidate learning content in the extended combination of content of interest in the learning content distribution map, for example, Figure 3 In the figure, candidate learning content 1 is associated with candidate learning content 2, candidate learning content 3 and candidate learning content 4. It should be noted that since the association configuration between the candidate learning contents is determined based on the association relationship, the associated extended learning contents of the content extension combination of interest in the learning content distribution map are mapped into vector space, so that the obtained learning content extension vector is integrated in the subsequent processing, thereby improving the representation ability of the vector finally mined, thereby ensuring the accuracy of the obtained content recommendation results.

[0108] It can be understood that, in the above step S131c, the specific manner of associating and merging each candidate learning content semantic vector, each candidate learning content distribution vector and the learning content extension vector is not limited. For example, in a specific implementation, the above step S131b may include the following specific sub-steps that can be implemented:

[0109] First, the content association merging subnetwork included in the target learning content recommendation network can be used to perform vector merging operations on each candidate learning content semantic vector and the corresponding candidate learning content distribution vector, and output the corresponding content semantic distribution merged vector; exemplarily, in order to improve the efficiency of the vector merging operation, the candidate learning content semantic vector and the corresponding candidate learning content distribution vector can be spliced ​​or superimposed, and then the spliced ​​or superimposed vector can be convoluted, and pooling and full connection can also be performed, so as to obtain a content semantic distribution merged vector that carries both the candidate learning content semantics and the distribution relationship;

[0110] Secondly, based on the corresponding learning content extension vector, a focused merging operation can be performed on each of the content semantic distribution merged vectors to form corresponding learning content global vectors, and based on each of the learning content global vectors, a learning content global vector set corresponding to the interested content extension combination can be determined; wherein, the focused merging operation refers to mining out relevant information in the learning content extension vector and integrating it into the content semantic distribution merged vector, so that the representation ability of the obtained learning content global vector is better, and the specific process of the focused merging operation can refer to the relevant description below, and can refer to the process of obtaining the extension combination correlation parameters, so that the learning content global vectors corresponding to each candidate learning content in the interested content extension combination are combined to form a corresponding learning content global vector set.

[0111] It can be understood that in the above-mentioned step S132, the specific method of performing focusing and merging operations on the learning content global vector set and the learning content extension vector corresponding to the content extension combination of interest (i.e., performing focusing and merging operations on the learning content global vectors in the learning content global vector set based on the learning content extension vector) is not limited and can be selected according to actual needs.

[0112] For example, in a specific implementation, in order to enable the focus merging operation to fully integrate the relevant information in the learning content extension vector with the semantic information in the learning content global vector, and obtain a learning content focus merging vector with higher accuracy in information representation, the above step S132 may further include the following contents (such as Figure 4 ):

[0113] First, for each of the at least one extended combination of content of interest, the first spatial mapping parameter, the second spatial mapping parameter and the third spatial mapping parameter included in the target learning content recommendation network are used to spatially map each of the learning content global vectors and the learning content extended vector in the learning content global vector set corresponding to the extended combination of content of interest, and the first spatial mapping result, the second spatial mapping result and the third spatial mapping result corresponding to each of the learning content global vectors are output, and, based on the first spatial mapping result, the second spatial mapping result and the third spatial mapping result, the extended combination association parameter corresponding to the learning content global vector is merged and output to obtain the extended combination corresponding to each of the learning content global vectors in the extended combination of content of interest. Correlation parameters; for example, for the learning content global vector 1, the learning content global vector 1 can be multiplied by the first space mapping parameter to obtain the first space mapping result, and the learning content extension vector can be multiplied by the second space mapping parameter and the third space mapping parameter respectively to obtain the second space mapping result and the third space mapping result. After that, the first space mapping result and the second space mapping result can be multiplied, so that a similarity relationship parameter can be obtained. Then, based on the similarity relationship parameter, the third space mapping result is weighted to obtain the extended combination correlation parameter corresponding to the learning content global vector 1. Based on the same method, the extended combination correlation parameter corresponding to the learning content global vector 2 and the extended combination correlation parameter corresponding to the learning content global vector 3 can also be determined;

[0114] Secondly, based on the extended combination correlation parameters corresponding to each of the learning content global vectors in an extended combination of the content of interest and each of the learning content global vectors, the learning content focus merged vector of the extended combination of the content of interest is merged and output. For example, the learning content global vector 1 and the corresponding extended combination correlation parameters can be superimposed to obtain a first vector; the learning content global vector 2 and the corresponding extended combination correlation parameters can be superimposed to obtain a second vector; the learning content global vector 3 and the corresponding extended combination correlation parameters can be superimposed to obtain a third vector; then, the first vector, the second vector and the third vector can be spliced ​​and then pooled to obtain the corresponding learning content focus merged vector.

[0115] For another example, in another alternative implementation, in order to ensure that the focus merging operation can fully integrate the relevant information in the learning content extension vector with the semantic information in the learning content global vector, and obtain a learning content focus merging vector with higher accuracy in information representation, the above step S132 may further include the following contents:

[0116] First, for each learning content global vector corresponding to each of the at least one interested content extension combination, the learning content global vector and the learning content extension vector corresponding to the interested content extension combination are multiplied, and the importance coefficient corresponding to the learning content global vector is output, wherein the importance coefficient refers to the importance relative to the correlation with the learning content extension vector, such as the higher the correlation, the greater the importance, so that the importance coefficient corresponding to the learning content global vector 1, the importance coefficient corresponding to the learning content global vector 2, the importance coefficient corresponding to the learning content global vector 3, etc. can be obtained;

[0117] Then, for each of the at least one content extension combination of interest, based on the importance coefficient corresponding to each learning content global vector in the content extension combination of interest, a weighted summation calculation is performed on each learning content global vector in the content extension combination of interest, and the learning content focus merging vector of the content extension combination of interest is output; that is, the corresponding importance coefficient can be used as the corresponding weighting coefficient to perform a weighted summation calculation, so that information aggregation based on the correlation with the learning content extension vector can be achieved, so that the merging effect of the learning content extension vector is better, and the semantic information of the learning content global vector itself will not be lost.

[0118] It can be understood that, in the above step S133, the specific manner of performing content recommendation processing on the learning content focus merging vector corresponding to the interesting content extension combination is not limited. For example, in a specific implementation, in order to ensure the accuracy of content recommendation processing, the above step S133 may further include the following specific implementation process:

[0119] First, a local learning content distribution map corresponding to the at least one extended combination of content of interest may be extracted from the learning content distribution map, wherein the local learning content distribution map includes each candidate learning content in the at least one extended combination of content of interest and the corresponding association relationship, such as Figure 4 In the example, candidate learning content 1 and candidate learning content 2 have an association relationship, while candidate learning content 1 and candidate learning content 5 do not have an association relationship;

[0120] Secondly, a content mining operation can be performed on the local learning content distribution map to output a corresponding local map semantic vector, wherein the local map semantic vector is used to reflect the semantic information of each candidate learning content and the semantic information of the corresponding association relationship in the local learning content distribution map; illustratively, the candidate learning content and the corresponding association relationship included in the local learning content distribution map can be subjected to word embedding processing, and then the vectors obtained by the word embedding processing are spliced, and then the obtained spliced ​​vectors are convolved to obtain the local map semantic vector;

[0121] Then, based on the number of the at least one content of interest extension combination, a corresponding number of random noise vectors can be generated, wherein the size of each of the random noise vectors is consistent with the size of the local graph semantic vector, and for any of the random noise vectors, the number of non-zero parameters in the random noise vector is less than the number of zero parameters (it should be noted that the number of non-zero parameters can be much smaller than the number of zero parameters to avoid introducing too much noise to affect the expression of features), and the number of non-zero parameters between any two of the random noise vectors is the same, the distribution shape is the same, and the distribution positions are not exactly the same. For example, an all-zero vector can be generated, and then, based on a rectangular frame, multiple different positions of the all-zero vector are framed, and the parameters of the framed positions are set to non-zero parameters, respectively, so as to form corresponding random noise vectors; in addition, the non-zero parameters between different random noise vectors can be the same or different;

[0122] Further, each of the random noise vectors and the local graph semantic vector may be respectively superimposed to form at least one corresponding local graph noise vector, and the at least one local graph noise vector may be assigned to the at least one content of interest extension combination, so that a one-to-one correspondence is formed between the at least one local graph noise vector and the at least one content of interest extension combination; illustratively, the assignment may be random;

[0123] Finally, for each of the at least one interest content extension combination, based on the local graph noise vector corresponding to the interest content extension combination, a focusing operation is performed on the learning content focus merging vector corresponding to the interest content extension combination to form a corresponding target vector (for example, based on a set of first spatial mapping parameters, second spatial mapping parameters, and third spatial mapping parameters different from the aforementioned parameters, the local graph noise vector (multiplied by the first spatial mapping parameter) and the learning content focus merging vector (multiplied by the second spatial mapping parameter and the third spatial mapping parameter, respectively) are spatially mapped, and then the first spatial mapping result is calculated. and the dot product between the second space mapping result, and based on the dot product, the third space mapping result is weighted to obtain a target vector), and the target learning content recommendation network is used to perform content recommendation processing on the target vector to form a content recommendation result corresponding to the extended combination of content of interest; exemplarily, the fully connected subnetwork in the target learning content recommendation network can be used to perform full connection processing on the target vector, and then the classification function (such as softmax function, etc.) or regression function in the target learning content recommendation network is used to classify or regress the vector output by the full connection processing to obtain the content recommendation result corresponding to the extended combination of content of interest.

[0124] It can be understood that, in another specific implementation, in order to ensure the accuracy of content recommendation processing, the above step S133 may further include the following specific implementation process:

[0125] First, a local learning content distribution map corresponding to the at least one extended combination of content of interest may be extracted from the learning content distribution map, wherein the local learning content distribution map includes each candidate learning content in the at least one extended combination of content of interest and the corresponding association relationship, such as Figure 4 In the example, candidate learning content 1 and candidate learning content 2 have an association relationship, while candidate learning content 1 and candidate learning content 5 do not have an association relationship;

[0126] Secondly, a content mining operation can be performed on the local learning content distribution map to output a corresponding local map semantic vector, wherein the local map semantic vector is used to reflect the semantic information of each candidate learning content and the semantic information of the corresponding association relationship in the local learning content distribution map; illustratively, the candidate learning content and the corresponding association relationship included in the local learning content distribution map can be subjected to word embedding processing, and then the vectors obtained by the word embedding processing are spliced, and then the obtained spliced ​​vectors are convolved to obtain the local map semantic vector;

[0127] Then, multiple random noise vectors can be generated (the number can be configured according to actual needs), wherein the size of each of the random noise vectors is consistent with the size of the local graph semantic vector, and for any of the random noise vectors, the number of non-zero parameters in the random noise vector is less than the number of zero parameters (it should be noted that the number of non-zero parameters can be much smaller than the number of zero parameters to avoid introducing too much noise affecting the expression of features), and the number of non-zero parameters between any two of the random noise vectors is the same, the distribution shape is the same, and the distribution positions are not exactly the same. For example, an all-zero vector can be generated, and then, based on a rectangular frame, multiple different positions of the all-zero vector are selected, and the parameters of the selected positions are set to non-zero parameters, thereby forming corresponding random noise vectors respectively; in addition, the non-zero parameters between different random noise vectors can be the same or different;

[0128] For each of the at least one interest content extension combination, focusing operations are performed on the learning content focus merging vectors corresponding to the interest content extension combination in sequence based on the multiple random noise vectors to obtain a target vector corresponding to the interest content extension combination, wherein for the first random noise vector, a focusing operation is performed on the learning content focus merging vector based on the random noise vector to obtain an intermediate vector corresponding to the random noise vector, and for each random noise vector other than the first random noise vector, a focusing operation is performed on the superimposed vector of the learning content focus merging vector and the intermediate vector corresponding to the previous random noise vector based on the random noise vector to obtain an intermediate vector corresponding to the random noise vector, and the target vector is a superposition result between the intermediate vector corresponding to the last random noise vector and the learning content focus merging vector;

[0129] Finally, for each of the at least one content extension combination of interest, the target learning content recommendation network is used to perform content recommendation processing on the target vector corresponding to the content extension combination of interest, so as to form a content recommendation result corresponding to the content extension combination of interest; exemplarily, the fully connected subnetwork in the target learning content recommendation network can be used to perform full connection processing on the target vector, and then the classification function (such as softmax function, etc.) or regression function in the target learning content recommendation network is used to classify or regress the vector output by the full connection processing, so as to obtain the content recommendation result corresponding to the content extension combination of interest.

[0130] It can be understood that, in a specific implementation manner, in order to enable the above-mentioned target learning content recommendation network to reliably implement the above-mentioned mining and analysis processing, before performing the above-mentioned steps, corresponding training can also be performed first, that is, the artificial intelligence-based learning content recommendation method can also include the step of training to form the target learning content recommendation network, or the above-mentioned step S130 can include:

[0131] First, the candidate learning content recommendation network can be updated and optimized based on multiple training learning content data clusters, and when the update optimization of the candidate learning content recommendation network matches the pre-configured update optimization condition (for example, the network optimization loss of the candidate learning content recommendation network converges or the number of update optimizations reaches a threshold), based on the current candidate learning content recommendation network, a target learning content recommendation network is determined (for example, the current candidate learning content recommendation network can be directly used as the target learning content recommendation network), wherein, during the update optimization process:

[0132] For each loaded training learning content data cluster, the candidate learning content recommendation network is used to analyze the content recommendation results corresponding to each training learning content data in the currently loaded training learning content data cluster (the specific analysis process is as described above, and the specific content of the training learning content data can refer to all the parameters included in the interest content extension combination in the previous text), and based on the content recommendation results and the actual recommendation level corresponding to the training learning content data (it can be understood that the actual recommendation level can be obtained based on other reliable neural network model evaluations, or can be obtained based on the analysis and evaluation of the corresponding experts from the perspectives of the overall comprehensiveness and relevance of the learning content. In this way, in each application, it is not necessary to have the corresponding experts perform the analysis, thereby reducing the labor cost to a large extent and improving the processing efficiency), determine the corresponding network optimization loss (such as the mean square error), and, along the direction of reducing the network optimization loss, update and optimize the network parameters of the candidate learning content recommendation network (the network parameters include the first space mapping parameters, the second space mapping parameters and the third space mapping parameters, etc.) wherein each of the training learning content data corresponds to a training interest content extension combination.

[0133] In summary, the artificial intelligence-based learning content recommendation method and system provided by the present invention first construct a learning content distribution map; secondly, based on two interested learning contents, the interested content is expanded in the learning content distribution map to form at least one interested content expansion combination; then, using the target learning content recommendation network, content recommendation processing is performed on each interested content expansion combination respectively to obtain corresponding content recommendation results; finally, based on the corresponding content recommendation results, a target interested content expansion combination is determined in at least one interested content expansion combination, and based on the learning content included in the target interested content expansion combination, a learning content recommendation operation is performed on the target learning user. Based on the above method, since a learning content distribution map will be constructed, the reliability of the expansion of the content of interest based on the learning content distribution map is higher (that is, the association relationship in the learning content distribution map is fully utilized for expansion, so that compared with the conventional arbitrary expansion, it can have higher reliability), and, after obtaining at least one content of interest, the recommendation degree will be evaluated based on the target learning content recommendation network, so that the target content of interest expansion combination can be determined based on the obtained content recommendation results. In this way, the reliability of the learning content recommended to the target learning user can be further improved (compared with the conventional arbitrary selection of learning content, it can have higher reliability), thereby improving the problem of low reliability of learning content recommendation in the prior art, so that learning resources can be effectively utilized and the problem of resource waste can be avoided. Therefore, it also has a high application value in practical applications.

[0134] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0135] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0136] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention is essentially or partly contributed to the prior art or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk. It should be noted that, in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements inherent to such a process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0137] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A learning content recommendation method based on artificial intelligence, characterized in that: include: Based on the multiple candidate learning contents, a corresponding learning content distribution map is constructed, wherein the learning content distribution map is composed of the multiple candidate learning contents and the association relationships between the multiple candidate learning contents, and in the learning content distribution map, the name of the map node is the name of the corresponding candidate learning content, and the attribute information of the map node is the corresponding candidate learning content; Determine two candidate learning contents from the plurality of candidate learning contents as the learning contents of interest to the target learning user, and based on the two learning contents of interest, perform content of interest expansion in the learning content distribution map to form at least one content of interest expansion combination, wherein each of the content of interest expansion combinations includes the two learning contents of interest; Using the target learning content recommendation network, each of the interest content extension combinations is subjected to content recommendation processing to obtain corresponding content recommendation results, wherein the target learning content recommendation network is a neural network formed through network optimization, and the content recommendation results are used to reflect the recommendation degree of the learning content included in the corresponding interest content extension combination; Based on the corresponding content recommendation result, in the at least one extended combination of content of interest, a target extended combination of content of interest is determined, and based on the learning content included in the target extended combination of content of interest, a learning content recommendation operation is performed on the target learning user; The step of determining two candidate learning contents from the plurality of candidate learning contents as the learning contents of interest to the target learning user, and expanding the contents of interest in the learning content distribution map based on the two learning contents of interest to form at least one expanded combination of contents of interest includes: In response to a selection operation performed by a target learning user, two candidate learning contents are determined from the plurality of candidate learning contents as learning contents of interest to the target learning user, or, based on historical learning data of the target learning user, two candidate learning contents are determined from the plurality of candidate learning contents as learning contents of interest to the target learning user; Determine one of the two learning contents of interest as a first learning content of interest, and determine the other of the two learning contents of interest as a second learning content of interest; Starting from the first learning content of interest, wandering in the learning content distribution map until wandering to the second learning content of interest, so as to form at least one wandering sequence, wherein in each wandering sequence, no two identical candidate learning contents or map nodes are included, and when multiple wandering sequences are formed, every two wandering sequences are at least partially different; At least one content-of-interest extension combination is formed based on the at least one walking sequence.

2. The method for recommending learning content based on artificial intelligence according to claim 1, characterized in that: The step of constructing a corresponding learning content distribution map based on multiple candidate learning contents includes: Based on the multiple candidate learning contents, determining a corresponding multiple graph nodes, wherein there is a one-to-one correspondence between the multiple graph nodes and the multiple candidate learning contents; For every two graph nodes, the semantic similarity of the two candidate learning contents corresponding to the two graph nodes is calculated to obtain the semantic similarity corresponding to the two graph nodes; For every two graph nodes, the overlap degree of the learning user sets corresponding to the two graph nodes is calculated to obtain the user overlap degree corresponding to the two graph nodes, wherein the learning user set refers to the set of historical learning users who have studied or recommended in history; For every two graph nodes, based on the semantic similarity and user overlap corresponding to the two graph nodes, determine whether there is a correlation between the two graph nodes, and when there is no correlation between the two graph nodes, determine not to associate the two graph nodes. Alternatively, when there is a correlation between the two graph nodes, associate the two graph nodes to form a corresponding learning content distribution map, wherein in the learning content distribution map, there is an associated line segment between the two associated graph nodes.

3. The method for recommending learning content based on artificial intelligence according to claim 1, characterized in that: The step of using the target learning content recommendation network to perform content recommendation processing on each of the interest content extension combinations to obtain corresponding content recommendation results includes: Using a target learning content recommendation network, performing a content mining operation on the at least one interest content extension combination to form a learning content global vector set and a learning content extension vector for each interest content extension combination in the at least one interest content extension combination, wherein the learning content global vector in the learning content global vector set serves as a semantic vector of each candidate learning content in the corresponding interest content extension combination, and the learning content extension vector serves as a semantic vector of an associated extended learning content of each candidate learning content in the corresponding interest content extension combination in the learning content distribution map; For each of the at least one interest content extension combination, using the target learning content recommendation network, a learning content global vector set and a learning content extension vector corresponding to the interest content extension combination are focused and merged to form a learning content focused merged vector corresponding to the interest content extension combination; For each of the at least one interest content extension combination, the target learning content recommendation network is used to perform content recommendation processing on the learning content focus merging vector corresponding to the interest content extension combination to form a content recommendation result corresponding to the interest content extension combination.

4. The method for recommending learning content based on artificial intelligence according to claim 3, characterized in that: The step of using the target learning content recommendation network to perform content mining operations on the at least one interesting content extension combination to form a learning content global vector set and a learning content extension vector for each interesting content extension combination in the at least one interesting content extension combination includes: For any of the interest content extension combinations, the following operations are performed: Using the content mining subnetwork included in the target learning content recommendation network to perform content mining operations, forming candidate learning content semantic vectors and candidate learning content distribution vectors corresponding to each candidate learning content in the interest content extension combination, and forming a learning content extension vector corresponding to the interest content extension combination, wherein the candidate learning content semantic vector is used to reflect the semantic information of the candidate learning content itself, and the candidate learning content distribution vector is used to reflect the distribution relationship information of the candidate learning content in the interest content extension combination; By utilizing the content association merging subnetwork included in the target learning content recommendation network, each candidate learning content semantic vector, each candidate learning content distribution vector and the learning content extension vector are associated and merged to form a global learning content vector set corresponding to the interested content extension combination.

5. The method for recommending learning content based on artificial intelligence as claimed in claim 4, characterized in that: The step of performing content mining operations using the content mining subnetwork included in the target learning content recommendation network to form candidate learning content semantic vectors and candidate learning content distribution vectors corresponding to each candidate learning content in the content extension combination of interest, and forming a learning content extension vector corresponding to the content extension combination of interest, includes: Using the content mining unit in the content mining subnetwork included in the target learning content recommendation network, each candidate learning content in the extended combination of interested content and the distribution relationship information of each candidate learning content are mapped into a vector space to form a candidate learning content semantic vector corresponding to each candidate learning content and a candidate learning content distribution vector corresponding to each candidate learning content; Utilizing the extended combination mining unit in the content mining subnetwork, the associated extended learning contents of the extended content combination of interest in the learning content distribution map are mapped into a vector space to form a learning content extension vector corresponding to the extended content combination of interest, wherein the associated extended learning contents include other candidate learning contents of each associated configuration of the candidate learning contents in the extended content combination of interest in the learning content distribution map.

6. The method for recommending learning content based on artificial intelligence according to claim 4, characterized in that: The step of using the content association merging subnetwork included in the target learning content recommendation network to associate and merge each candidate learning content semantic vector, each candidate learning content distribution vector and the learning content extension vector to form a global learning content vector set corresponding to the interested content extension combination includes: Utilizing the content association merging subnetwork included in the target learning content recommendation network, respectively merging each candidate learning content semantic vector and the corresponding candidate learning content distribution vector, and outputting the corresponding content semantic distribution merging vector; Based on the corresponding learning content extension vectors, focus merging operations are performed on each of the content semantic distribution merge vectors to form corresponding learning content global vectors, and based on each of the learning content global vectors, a learning content global vector set corresponding to the interested content extension combination is determined.

7. The method for recommending learning content based on artificial intelligence as claimed in claim 3, characterized in that: The step of performing a focus merging operation on a learning content global vector set and a learning content extension vector corresponding to each of the at least one interesting content extension combination by using the target learning content recommendation network to form a learning content focus merging vector corresponding to the interesting content extension combination comprises: For each of the at least one extended combination of content of interest, use the first spatial mapping parameter, the second spatial mapping parameter and the third spatial mapping parameter included in the target learning content recommendation network to spatially map each of the global learning content vectors and the learning content extended vector in the global learning content vector set corresponding to the extended combination of content of interest, output the first spatial mapping result, the second spatial mapping result and the third spatial mapping result corresponding to each of the global learning content vectors, and, based on the first spatial mapping result, the second spatial mapping result and the third spatial mapping result, merge and output the extended combination association parameter corresponding to the global learning content vector, so as to obtain the extended combination association parameter corresponding to each of the global learning content vectors in the extended combination of content of interest; Based on the extended combination relevance parameter corresponding to each of the learning content global vectors in an extended combination of the content of interest and each of the learning content global vectors, a learning content focus merged vector of the extended combination of the content of interest is merged and output.

8. The method for recommending learning content based on artificial intelligence as claimed in claim 3, characterized in that: The step of performing content recommendation processing on the learning content focus merging vector corresponding to the at least one interested content extension combination by using the target learning content recommendation network to form a content recommendation result corresponding to the interested content extension combination for each interested content extension combination of the at least one interested content extension combination comprises: Extracting a local learning content distribution map corresponding to the at least one extended combination of content of interest from the learning content distribution map, wherein the local learning content distribution map includes each candidate learning content in the at least one extended combination of content of interest and a corresponding association relationship; Performing content mining operations on the local learning content distribution map, and outputting a corresponding local map semantic vector, wherein the local map semantic vector is used to reflect the semantic information of each candidate learning content in the local learning content distribution map and the semantic information of the corresponding association relationship; Based on the number of the at least one combination of content extensions of interest, a corresponding number of random noise vectors are generated, wherein the size of each of the random noise vectors is consistent with the size of the local graph semantic vector, and for any one of the random noise vectors, the number of non-zero parameters in the random noise vector is less than the number of zero parameters, and the number of non-zero parameters between any two of the random noise vectors is the same, the distribution shape is the same, and the distribution positions are not completely the same; Superimposing each of the random noise vectors and the local graph semantic vector respectively to form at least one corresponding local graph noise vector, and assigning the at least one local graph noise vector to the at least one content of interest extension combination, so that a one-to-one correspondence is formed between the at least one local graph noise vector and the at least one content of interest extension combination; For each of the at least one content extension combination of interest, based on the local graph noise vector corresponding to the content extension combination of interest, a focusing operation is performed on the learning content focus merging vector corresponding to the content extension combination of interest to form a corresponding target vector, and the target learning content recommendation network is used to perform content recommendation processing on the target vector to form a content recommendation result corresponding to the content extension combination of interest.

9. A learning content recommendation system based on artificial intelligence, characterized in that: It includes a processor and a memory, the memory is used to store a computer program, and the processor is used to execute the computer program to implement the artificial intelligence-based learning content recommendation method described in any one of claims 1 to 8.

Images