Agricultural knowledge recommendation method and device

Abstract

The present invention provides an agricultural knowledge recommendation method and device, which relates to the field of agricultural information technology. The method includes: obtaining basic attribute information and historical behavior data of a target user; determining the time subspace, regional subspace, preference subspace and interactive information subspace corresponding to the target user based on the basic attribute information and historical behavior data; obtaining agricultural knowledge recommendation results for the target user from a first agricultural knowledge graph based on the time subspace, regional subspace, preference subspace and interactive information subspace; wherein the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge. The agricultural knowledge recommendation method and device provided by the present invention can improve the accuracy, effectiveness and timeliness of recommending agricultural knowledge to agricultural science and technology personnel and farmers by comprehensively utilizing time and space, crop rotation, variety, user and agricultural knowledge interactive information, and performing knowledge reasoning on the agricultural knowledge graph containing multimodal agricultural knowledge, thereby improving the accuracy, effectiveness and timeliness of recommending agricultural knowledge to agricultural science and technology personnel and farmers.

Description

Technical Field

[0001] The present invention relates to the field of agricultural information technology, and in particular to an agricultural knowledge recommendation method and device. Background Art

[0002] In order to address the problems of poor acceptance of agricultural technology knowledge by farmers due to the delay in the delivery of agricultural experts / knowledge / technology to villages and rural areas, inaccurate matching of scientific and technological resources, and inconsistent technical service standards, there is an urgent need to innovate agricultural technology service methods and make the transfer of agricultural knowledge and technology smarter and simpler based on informatization.

[0003] In recent years, with the development of artificial intelligence technology, various forms of recommendation models for image and text semantic understanding have developed rapidly and made significant progress, including collaborative filtering-based recommendation, content (text, image)-based recommendation, hybrid model-based recommendation, etc. Among them, the intelligent recommendation algorithm based on knowledge graph can make full use of the graph structure database, build reliable entity relationship tuples, accurately mine the fine-grained cross-modal implicit semantics between entities and relationships, so as to accurately find the relationship path connecting entity pairs and realize high-precision recommendation of agricultural knowledge, playing an irreplaceable role in multimodal information processing and logical relationship expression methods.

[0004] However, due to the scattered distribution of agricultural data, the relevant technologies for agricultural knowledge recommendation based on knowledge graphs have defects such as poor accuracy and timeliness of knowledge services. Summary of the invention

[0005] The present invention provides an agricultural knowledge recommendation method and device, which are used to solve the defects of poor accuracy and timeliness of knowledge services in the prior art, and to improve the accuracy and effectiveness of agricultural data services for agricultural science and technology personnel and farmers in obtaining agricultural knowledge.

[0006] The present invention provides an agricultural knowledge recommendation method, comprising:

[0007] Obtain basic attribute information and historical behavior data of target users;

[0008] Based on the basic attribute information and the historical behavior data, determining the time subspace, geographical subspace, preference subspace and interactive information subspace corresponding to the target user;

[0009] Based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, obtaining agricultural knowledge recommendation results for the target user from the first agricultural knowledge graph;

[0010] Among them, the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0011] According to an agricultural knowledge recommendation method provided by the present invention, before obtaining the agricultural knowledge recommendation result for the target user from the first agricultural knowledge graph based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, the method further includes:

[0012] According to the semantics of the entity, the second agricultural knowledge graph is subjected to dimensionality reduction processing and multimodal information fusion to obtain the first agricultural knowledge graph;

[0013] Among them, the second agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0014] According to an agricultural knowledge recommendation method provided by the present invention, the agricultural knowledge recommendation result for the target user is obtained from the first agricultural knowledge graph based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, including:

[0015] Determine candidate paths in the first agricultural knowledge graph based on the time subspace, the geographical subspace, the preference subspace, and the interactive information subspace;

[0016] Based on the candidate paths, the agricultural knowledge recommendation results are obtained.

[0017] According to an agricultural knowledge recommendation method provided by the present invention, the time subspace, geographical subspace, preference subspace and interactive information subspace corresponding to the target user are determined based on the basic attribute information and the historical behavior data, including:

[0018] Based on the current time information in the basic attribute information, obtaining a time subspace matrix for representing the time subspace;

[0019] Based on the region information of the target user in the basic attribute information, obtaining a region subspace matrix for representing the region subspace;

[0020] Based on the crop rotation information, farming information, agricultural condition information and content information in the basic attribute information, obtaining a preference subspace matrix for representing the preference subspace;

[0021] Based on the number of clicks and browsing time of the target user on each piece of knowledge in the historical behavior data, an interactive information subspace matrix for characterizing the interactive information subspace is obtained.

[0022] According to an agricultural knowledge recommendation method provided by the present invention, after obtaining the agricultural knowledge recommendation result for the target user from the first agricultural knowledge graph based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, the method further includes:

[0023] The agricultural knowledge recommendation result is evaluated to obtain an evaluation score of the agricultural knowledge recommendation result.

[0024] According to an agricultural knowledge recommendation method provided by the present invention, multimodal information fusion is performed on the second agricultural knowledge graph according to the semantics of the entity, including:

[0025] Entities with the same semantics in the second agricultural knowledge graph are fused.

[0026] The present invention also provides an agricultural knowledge recommendation device, comprising:

[0027] Data acquisition module, used to obtain basic attribute information and historical behavior data of target users;

[0028] A subspace determination module, used to determine the time subspace, geographical subspace, preference subspace and interaction information subspace corresponding to the target user based on the basic attribute information and the historical behavior data;

[0029] A knowledge recommendation module, configured to obtain agricultural knowledge recommendation results for the target user from the first agricultural knowledge graph based on the time subspace, the geographical subspace, the preference subspace and the interactive information subspace;

[0030] Among them, the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0031] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the agricultural knowledge recommendation methods described above is implemented.

[0032] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the agricultural knowledge recommendation method as described in any one of the above is implemented.

[0033] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the agricultural knowledge recommendation method as described above is implemented.

[0034] The agricultural knowledge recommendation method and device provided by the present invention, after determining the time subspace, regional subspace, preference subspace and interactive information subspace corresponding to the target user based on the basic attribute information and historical behavior data of the target user, comprehensively utilizes time and space, crop rotation, variety, target user and agricultural knowledge interactive information, and performs knowledge reasoning on the first agricultural knowledge graph containing multimodal agricultural knowledge, which can effectively improve the comprehensive utilization rate of text, image, spatiotemporal information and human-computer interactive information in the agricultural intelligent knowledge service platform, ensure that agricultural knowledge reaches agricultural technicians and farmers quickly and accurately, support fast and accurate services, thereby improving production quality and efficiency, and with the continuous expansion and enrichment of data, realize more accurate and comprehensive knowledge hybrid recommendation, which can improve the accuracy, effectiveness and timeliness of recommending agricultural knowledge to agricultural science and technology personnel and farmers. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 This is one of the flow charts of the agricultural knowledge recommendation method provided by the present invention;

[0037] Figure 2 is a schematic diagram of a subspace matrix in the agricultural knowledge recommendation method provided by the present invention;

[0038] Figure 3 This is the second flow chart of the agricultural knowledge recommendation method provided by the present invention;

[0039] Figure 4 It is a flow chart for evaluating the agricultural knowledge recommendation results in the agricultural knowledge recommendation method provided by the present invention;

[0040] Figure 5 It is a structural diagram of the service architecture in the agricultural knowledge recommendation method provided by the present invention;

[0041] Figure 6 It is a structural schematic diagram of the agricultural knowledge recommendation device provided by the present invention;

[0042] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. 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.

[0044] In the description of the invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0045] It should be noted that, relying on the development dividends of the new generation of artificial intelligence technology, machines can have human intelligence, and through advanced speech recognition models and graphical semantic analysis algorithms, intelligent knowledge services can be built to "replace" experts in agricultural technical guidance, thereby reducing the intensity of expert guidance work, expanding the scope of influence of professional knowledge, and improving the standardization level of agricultural knowledge services.

[0046] Among the related technologies, agricultural intelligent knowledge services emerged from the construction of expert systems. Currently, there are relatively complete expert system knowledge recommendation services for major crops, vegetables, fruit trees, etc. In recent years, with the development of artificial intelligence technology, various forms of recommendation models for image and text semantic understanding have developed rapidly and made significant progress, including typical methods such as collaborative filtering-based recommendation, content (text, image)-based recommendation, and hybrid model-based recommendation.

[0047] Thanks to the graphical associative structure of entities, relationships, and attributes in the knowledge graph, the knowledge graph has a natural advantage in the logical expression of the long chain of knowledge services, and has constructed vertical field knowledge graphs for pests and diseases, varieties, cultivation, market prices and other aspects of various crops.

[0048] Among them, intelligent recommendation algorithms and intelligent question-answering models based on knowledge graphs have attracted much attention, especially playing an irreplaceable role in multimodal information processing and logical relationship expression methods. Therefore, we should make full use of the graph structure database, build reliable entity relationship tuples, and accurately mine the fine-grained cross-modal implicit semantics between entities and relationships, so as to accurately find the relationship path connecting entity pairs and realize high-precision recommendation of agricultural knowledge.

[0049] In the related technology, a multimodal knowledge graph can be constructed based on the LEBERT model, and an entity-relationship joint extraction model is trained to realize a multimodal knowledge graph that can automatically extract relationship triplets, which saves the time and effort of knowledge graph construction to a certain extent; the entity matching of the knowledge graph can also be realized through the method of multimodal pattern migration, and the corresponding semantic templates and visual patterns are extracted from different categories of head entities and migrated to the image acquisition process of the same type of non-head entities. The search engine keywords are constructed through semantic templates, and the retrieval results are denoised, which has great advantages in the accuracy of entity corresponding images; the knowledge graph and the image feature KG-I recommendation model can also be integrated, and visual embedding, knowledge embedding, and structural embedding methods are added to mine implicit feedback information between target user projects, and the Deep Walk model is used to capture the spatial structure method. The Ripple-Net mining knowledge graph expression idea effectively alleviates the data coefficient situation and solves the problem of a single knowledge graph ignoring rich knowledge information to a certain extent.

[0050] The above methods rely on deep learning models in the construction of multimodal knowledge graphs, and have initially realized the automated construction of lightweight data samples, which has accelerated the speed of agricultural knowledge generation and logical structure construction to a certain extent. However, the adaptability, composition accuracy and real-time service support effects for massive agricultural feature corpora need further exploration.

[0051] Although the above methods have achieved recommendations in both text and image modes based on knowledge graph recommendation technology, agricultural data represented by agricultural conditions often carry spatiotemporal information characteristics, including longitude and latitude, time, etc., as well as crop physiological information and growth law characteristics, including varieties, crops, agronomy, growth environment, etc. How to use this information to integrate key information such as varieties, crops, agronomy, time, space, etc. for agricultural production knowledge recommendations with basic factors in the model to form a recommendation service with agronomic characteristics is to improve the accuracy and effectiveness of agricultural data services for agricultural science and technology personnel and farmers in acquiring agricultural knowledge, and to make full use of agricultural multimodal data is an urgent problem to be solved.

[0052] In this regard, the present invention provides an agricultural knowledge recommendation method. The agricultural knowledge recommendation method provided by the present invention is aimed at the knowledge acquisition and utilization of agricultural science and technology personnel and farmers, and considers the comprehensive utilization of time and space, crop rotation, variety, target users and agricultural knowledge interaction information, and constructs a multimodal knowledge graph that integrates the above knowledge clues to express agricultural knowledge in graph structured data and entity-relationship tuples. Then, through the relationship between entities, relationships and attributes in the graph, agricultural knowledge is quickly recommended in terms of crop varieties, plant protection, agricultural conditions, nutrition, etc., relying on agricultural entity relationship extraction and fusion methods.

[0053] The agricultural knowledge recommendation method provided by the present invention is to support multi-source and multi-modal integrated applications such as plant and animal growth monitoring, agricultural knowledge graph evolution, and agricultural situation inversion. For the intelligent recommendation of agricultural knowledge, agricultural multimodal knowledge graphs are constructed respectively. Based on the characteristics of agricultural multi-source heterogeneous data combined with deep learning algorithms, automatic extraction of knowledge of different modalities is realized. User-knowledge collaborative filtering is performed with the help of subspace mapping methods, and the method of mining implicit knowledge is used to realize intelligent recommendation of agricultural knowledge. Subsequently, research on the knowledge recommendation service support architecture is carried out to support the rapid integration and iterative growth of massive agricultural knowledge and solve the problems of scattered systems and low data utilization.

[0054] Figure 1 This is one of the flow charts of the agricultural knowledge recommendation method provided by the present invention. Figure 1 The agricultural knowledge recommendation method of the present invention is described. Figure 1 As shown, the method includes: step 101, obtaining basic attribute information and historical behavior data of the target user.

[0055] It should be noted that the execution subject of the embodiment of the present invention is an agricultural knowledge recommendation device.

[0056] Specifically, the target user is the recommendation object of the agricultural knowledge recommendation method provided by the present invention, that is, the agricultural knowledge recommendation method provided by the present invention can recommend agricultural knowledge to the target user.

[0057] The basic attribute information of the target user may include current time information (such as timestamp), the region information where the target user is located, crop rotation information, farming information, agricultural information and content information, etc.

[0058] Optionally, the basic attribute information of the target user may also include at least one of the target user's industry, related agricultural crop varieties, gender, age, education level, and occupation (eg, grower, plant protection expert, or agricultural machinery operator, etc.).

[0059] The historical behavior data of the target user may include the number of clicks and browsing time of each piece of knowledge by the target user on the knowledge platform.

[0060] Optionally, the historical behavior data of the target user may further include at least one of the time when the knowledge was clicked, the category of the knowledge clicked, the reading dwell time, whether there is a comment, whether the user asked the question himself, and whether the question was adopted.

[0061] The embodiments of the present invention can obtain the basic attribute information and historical behavior data of the target user in a variety of ways. For example, the basic attribute information of the target user can be obtained based on the input of the target user; for another example, the historical behavior data of the target user can be obtained by data query. The embodiments of the present invention do not limit the specific method of obtaining the basic attribute information and historical behavior data of the target user.

[0062] Step 102: Based on the basic attribute information and the historical behavior data, determine the time subspace, geographical subspace, preference subspace and interactive information subspace corresponding to the target user.

[0063] Specifically, after obtaining the target user's basic attribute information and historical behavior data, the target user's time characteristics, regional characteristics, preference characteristics that characterize the target user's preferences, and interaction characteristics that characterize the interaction between the target user and the knowledge platform can be extracted based on the target user's basic attribute information and historical behavior data. Furthermore, based on the target user's time characteristics, regional characteristics, preference characteristics, and interaction characteristics, a corresponding time subspace, regional subspace, preference subspace, and interaction information subspace can be formed respectively.

[0064] As an optional embodiment, based on the basic attribute information and the historical behavior data, obtaining the time subspace, the regional subspace, the preference subspace and the interaction information subspace includes: based on the current time information in the basic attribute information, obtaining a time subspace matrix for characterizing the time subspace;

[0065] Based on the region information of the target user in the basic attribute information, a region subspace matrix for representing the region subspace is obtained;

[0066] Based on the crop rotation information, farming information, agricultural condition information and content information in the basic attribute information, a preference subspace matrix for representing the preference subspace is obtained;

[0067] Based on the number of clicks and browsing time of target users on each piece of knowledge in the historical behavior data, an interactive information subspace matrix for characterizing the interactive information subspace is obtained.

[0068] Specifically, in the embodiment of the present invention, a subspace matrix can be constructed based on the basic attribute information and historical behavior data of the target user, and the time subspace, geographical subspace, preference subspace and interactive information subspace corresponding to the target user can be represented in the form of a matrix.

[0069] The interactive information subspace matrix G corresponding to the target user can be used to describe the interactive information between the target user and agricultural knowledge. The dimension of the interactive information subspace matrix G is M*N. ijIndicates whether farmer i interacts with agricultural knowledge j, 1 means interaction has occurred, and 0 means no interaction has occurred;

[0070] The time subspace matrix, geographical subspace matrix and preference subspace matrix corresponding to the target user can be used to describe the basic attribute information of the target user, such as time, space, intersection, habitat and other information.

[0071] It can be understood that the number of the interaction information subspace matrix, the time subspace matrix, the region subspace matrix and the preference subspace matrix corresponding to the target user can be one or more.

[0072] Figure 2 It is a schematic diagram of the subspace matrix in the agricultural knowledge recommendation method provided by the present invention. Figure 2 The i in is used to identify different users. Figure 2 The u in is used to identify different data intervals. For example, Figure 2 u1 in the middle regional subspace matrix can represent the first region, u2 can represent the second region, u3 can represent the third region, etc. User i1 takes the value of 1 under u1 and u3, and takes the value of 0 under u2, which can indicate that user i1 is in the first region and the third region, and user i1 is not in the second region. The first region and the third region can be in a containment relationship.

[0073] Step 103: Based on the time subspace, the geographical subspace, the preference subspace and the interactive information subspace, obtain agricultural knowledge recommendation results for the target user from the first agricultural knowledge graph.

[0074] Among them, the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0075] Specifically, the first agricultural knowledge graph may be a multimodal knowledge graph, and the data types may include at least two of text, images, audio, video, latitude and longitude, time, and interactive information (farmers and knowledge).

[0076] Optionally, the first agricultural knowledge graph may be a multimodal knowledge graph pre-constructed before step 102 , or may be constructed based on multimodal agricultural knowledge after step 102 .

[0077] Preferably, the first agricultural knowledge graph can be constructed based on multimodal agricultural knowledge before step 102.

[0078] As an optional embodiment, before obtaining the agricultural knowledge recommendation results for the target user from the first agricultural knowledge graph based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, the method also includes: performing dimensionality reduction processing and multimodal information fusion on the second agricultural knowledge graph according to the semantics of the entity to obtain the first agricultural knowledge graph; wherein the second agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0079] Figure 3 This is the second flow chart of the agricultural knowledge recommendation method provided by the present invention. Figure 3 As shown, in the embodiment of the present invention, a multimodal second agricultural knowledge graph can be constructed based on a massive multimodal agricultural knowledge construction corpus. The construction scope may include "crop varieties, cultivation patterns, pests and diseases, plant protection information, agricultural product sales, agricultural technical services" and other aspects. Data types may include text, images, longitude and latitude, time, and interactive information (farmers and knowledge), etc., to provide a basis for subsequent subspace construction.

[0080] When constructing the second agricultural knowledge graph, we can first construct agricultural knowledge entities and relationship triplets, configure attribute values, and annotate the head entity, tail entity and related information in the BIOES (head entity-middle-non-entity-tail entity-single character) method. Then, we use the ERNIE (knowledge enhanced semantic representation model) pre-training model + attention mechanism to perform multi-feature fusion named entity recognition. In addition to text, heterogeneous agricultural knowledge data such as pictures and images are located through URL (Uniform Resource Locator System), and supporting annotations are provided for picture, video, text data, as well as the contained spatiotemporal information and target user interaction information to construct a second agricultural knowledge graph containing multimodal information.

[0081] It can be understood that the second agricultural knowledge graph is a multimodal knowledge graph, and the data types may include at least two of text, images, audio, video, longitude and latitude, time and interactive information (farmers and knowledge).

[0082] It should be noted that the second agricultural knowledge graph contains multimodal data corpus information. The processing of multimodal data corpus information directly affects the construction of the second agricultural knowledge graph, especially the semantically complex instances in the second agricultural knowledge graph, which have a significant impact on the accuracy of semantic analysis of the knowledge recommendation model and accumulate deviations into the knowledge recommendation results, causing the target users to receive incorrect feedback.

[0083] Optionally, multimodal cross-media data such as images, texts, spatiotemporal information, and regional information in the second agricultural knowledge graph may be screened, filtered, and classified.

[0084] Optionally, the ResNet and CNN models can be used to filter the image information in the second agricultural knowledge graph.

[0085] Optionally, the ERNIE pre-trained model can be used to screen the corpus information in the second agricultural knowledge graph.

[0086] Optionally, the existing entities in the second agricultural knowledge graph can be matched, projected, and entity-fused according to sequence information, and the fused information can also be embedded into the newly created entities, simplifying the number of entities and expanding the multi-modal semantic expression ability at the same time, providing a simplified path for subsequent knowledge reasoning and recommendation.

[0087] As an optional embodiment, multi-modal information fusion is performed on the second agricultural knowledge graph according to the semantics of the entities, including: fusing the entities with the same semantics in the second agricultural knowledge graph.

[0088] Specifically, multiple entities with the same semantics in the second agricultural knowledge graph can be fused into one entity, thereby simplifying the knowledge graph.

[0089] For example, the entity representing the text "tomato", the entity representing the picture of "tomato", and the entity representing the voice of "tomato" all have the semantics of "tomato", and the above three entities can be merged into one entity, which can be linked to the text "tomato", the picture of "tomato", and the voice of "tomato".

[0090] For example, the entity representing the Chinese text "pig" and the entity representing the English text "pig" both have the semantics of "tomato", and the above two entities can be merged into one entity, which can be linked to the Chinese text "pig" and the English text "pig".

[0091] It should be noted that the embodiments of the present invention can also further explore the representation learning, knowledge reasoning, and recommendation modes of single-modal and multi-modal data in the agricultural knowledge graph on the basis of knowledge graph construction, spatial processing, and induction, realizing the dynamic expansion of knowledge graph entities and relationships and the automatic update of knowledge.

[0092] In the embodiment of the present invention, in view of the current situation that agricultural data is scattered and the accuracy and timeliness of knowledge services are poor, a semantic expression and evaluation method of multidimensional subspace is proposed, and the semantics of Internet agricultural data are fully summarized and represented by multimodal knowledge graphs. The richness of semantic expression is expanded through dimensionality reduction processing of multimodal knowledge, and the efficiency of semantic processing is improved. Agricultural text, image, time, space, and user-agricultural knowledge resource interaction relationships are semantically embedded, and deep semantic expression of agricultural knowledge ontology information and interactive information is realized. Knowledge reasoning of the optimal path selection method is realized through cross-modal data entity mapping, residual connection and self-attention operation, providing effective support for intelligent recommendation services of multimodal agricultural knowledge. Effectively improve the comprehensive utilization rate of text, image, spatiotemporal information and human-computer interaction information in the agricultural intelligent knowledge service platform, ensure that agricultural knowledge reaches agricultural technicians and farmers quickly and accurately, support fast and accurate services, thereby improving production quality and efficiency, and with the continuous expansion and enrichment of data, more accurate and comprehensive knowledge hybrid recommendation is realized.

[0093] The embodiment of the present invention obtains the first agricultural knowledge graph by performing dimensionality reduction processing and multimodal information fusion on the second agricultural knowledge graph according to the semantics of the entity. The embodiment of the present invention can effectively utilize the multi-source heterogeneous massive information of agriculture, clean and screen the collected data, and realize "denoising" at the data level. The method of deep learning named entity recognition is used to automatically extract semantics from the data, and the basic unit of the "triplet" that constitutes the knowledge graph is used to connect the entities by utilizing the relationship between nodes to construct an agricultural knowledge graph with multiple modalities such as image, text, time, and space and basic reasoning. The paradigm of multimodal expected dimensionality reduction enables fewer types of entities to store a wider range of semantic information, which can save storage and computing resources, provide more efficient data reading, writing and expression capabilities, and provide a good basic framework for subsequent personalized agricultural knowledge recommendation, knowledge retrieval and other services.

[0094] After obtaining the first agricultural knowledge graph, multimodal agricultural knowledge reasoning can be performed based on the multimodal first agricultural knowledge graph, so that agricultural knowledge recommendation results for target users can be obtained from the first agricultural knowledge graph. On the basis of knowledge graph construction, spatial processing and induction, single-modal and multi-modal data can be further used in agricultural knowledge graphs to represent learning, knowledge reasoning and recommendation models, etc., to achieve dynamic expansion of knowledge graph entities and relationships and automatic updating of knowledge.

[0095] As an optional embodiment, based on the time subspace, the region subspace, the preference subspace and the interactive information subspace, obtaining the agricultural knowledge recommendation result for the target user from the first agricultural knowledge graph includes: determining a candidate path in the first agricultural knowledge graph based on the time subspace, the region subspace, the preference subspace and the interactive information subspace;

[0096] Specifically, the embodiment of the present invention can perform multimodal agricultural knowledge reasoning based on the first agricultural knowledge graph.

[0097] It should be noted that the candidate path search process in the embodiment of the present invention simulates the agricultural knowledge resource subspace matrix recommendation and analysis process in the related art.

[0098] When performing multimodal agricultural knowledge reasoning based on the first agricultural knowledge graph, path encoding can be used to define the initial point of the candidate path, and the self-attention operation of the residual connection can be used to add the path information, and the recommended information can be input into the embedding matrix to find the corresponding word embedding.

[0099] Based on the candidate paths, the agricultural knowledge recommendation results are obtained.

[0100] Specifically, the candidate paths are calculated through the attention mechanism and normalized and weighted, which can realize the representation learning of the reasoning process and the recommendation of agricultural knowledge in the graph mode.

[0101] Based on the traditional collaborative filtering method, the embodiment of the present invention adds multimodal information such as text, image, time, space, and interaction to improve the breadth of recommended information processing. This method takes into account the complex characteristics of agricultural data structure and the logical relationship between them, and relies on the constructed multimodal knowledge graph to further explore the semantic relationship and invisible association between images of different scales, different types of text, text-image, text-time series, etc. Taking the text-image processing method as an example, different types of agricultural knowledge reasoning optimal path selection algorithms are formed to obtain candidate paths.

[0102] The embodiment of the present invention uses a subspace mapping method to perform user-knowledge collaborative filtering and adopts a method of mining implicit knowledge to achieve intelligent recommendation of agricultural knowledge, which can achieve more accurate and comprehensive knowledge hybrid recommendation and improve the accuracy, effectiveness and timeliness of recommending agricultural knowledge to agricultural scientists and farmers.

[0103] The embodiment of the present invention determines the time subspace, regional subspace, preference subspace and interactive information subspace corresponding to the target user based on the basic attribute information and historical behavior data of the target user, and then comprehensively utilizes time and space, crop rotation, variety, target user and agricultural knowledge interaction information to perform knowledge reasoning on the first agricultural knowledge graph containing multimodal agricultural knowledge. It can effectively improve the comprehensive utilization rate of text, images, spatiotemporal information and human-computer interaction information in the agricultural intelligent knowledge service platform, ensure that agricultural knowledge reaches agricultural technicians and farmers quickly and accurately, support fast and accurate services, thereby improving production quality and efficiency, and with the continuous expansion and enrichment of data, more accurate and comprehensive knowledge hybrid recommendations can be achieved, which can improve the accuracy, effectiveness and timeliness of recommending agricultural knowledge to agricultural science and technology personnel and farmers.

[0104] As an optional embodiment, after obtaining the agricultural knowledge recommendation results for the target user from the first agricultural knowledge graph based on the time subspace, the geographical subspace, the preference subspace and the interactive information subspace, the method also includes: evaluating the agricultural knowledge recommendation results to obtain an evaluation score of the agricultural knowledge recommendation results.

[0105] Figure 4 : is a flow chart for evaluating the agricultural knowledge recommendation results in the agricultural knowledge recommendation method provided by the present invention. Figure 4 As shown, the evaluation process can adopt a multi-dimensional subspace mapping knowledge evaluation mechanism.

[0106] The subspace matrix maps the target farmer u and agricultural knowledge i in the form of vectors on the plane for processing. Four types of attention mechanisms, namely time, space, preference, and interaction information, are used to extract features from the data. The interaction matrix of all historical farmers and all elements of agricultural knowledge is embedded into a low-density vector space through a fully connected layer. The embedded vectors are obtained in the historical farmer-knowledge mapping layer and used as the input of the subspace vectors.

[0107] The attention mechanism is equipped with a pre-trained model. Since the preference subspace contains knowledge description information, the space contains a text processing hidden layer, which performs text segmentation and convolution classification to provide a more accurate basis for knowledge recommendation. After each subspace passes through the attention mechanism and hidden layer, a vector set is output. After the output, the weight allocation layer assigns values ​​again according to the weights. For example, some knowledge spaces have higher weights, and some knowledge times have higher weights. Finally, they enter the multi-layer perceptron to output the evaluation scores of the target users' agricultural knowledge recommendation results.

[0108] The embodiment of the present invention adopts multiple subspace mapping expressions and evaluation mechanisms based on collaborative filtering and content-based recommendation technology to target the complexity of agricultural knowledge. It not only considers the richness of semantic expressions of multi-source heterogeneous agricultural data, but also analyzes spatiotemporal information, target user interaction information, and multimodal knowledge information processed in the front-end links, and forms an evaluation mechanism for knowledge matching relevance of high-order semantics and service accuracy.

[0109] In view of the current situation that agricultural data is scattered and the accuracy and timeliness of knowledge services are poor, an embodiment of the present invention proposes a semantic expression and evaluation method of a multidimensional subspace, which uses a multimodal knowledge graph to fully summarize and represent the semantics of Internet agricultural data, and semantically embeds agricultural text, images, time, space, and user-agricultural knowledge resource interaction relationships, thereby realizing deep semantic expression of agricultural knowledge ontology information and interactive information, and realizing knowledge reasoning of the optimal path selection method through cross-modal data entity mapping, residual connection and self-attention operations, providing effective support for intelligent recommendation services of multimodal agricultural knowledge.

[0110] Figure 5 Schematic diagram of the structure of the service architecture in the agricultural knowledge recommendation method provided by the present invention. Figure 5 As shown, the service architecture as a whole can be divided into four layers: data layer, mapping space, model layer and output layer.

[0111] The data layer may include data covering time, region, variety, farming, crop rotation, agricultural conditions, target user interaction characteristics, resource libraries such as target users, varieties, agricultural conditions, logs, farming, and information such as the number of clicks on the target user's human-computer interaction with the platform and the target user's evaluation of knowledge resources. The above data and information can usually be stored in relational databases such as MySQL and graph databases such as Neo4j.

[0112] The mapping space may include spatial mapping information and weight information of feature data, providing a more accurate matching structure for data model construction in the recommendation process.

[0113] The model layer can include the attention mechanism of the knowledge graph construction process, collaborative filtering algorithm, ERNIE, attention mechanism, and other deep learning processing models and data fusion models.

[0114] The output layer mainly includes labeled data, recommendation scores, recommended resource lists, target user lists, etc., providing specific recommendation services.

[0115] It should be noted that the graph database and the relational database in the embodiment of the present invention can be converted into each other through entity mapping, and can provide real-time data docking capabilities for the online knowledge service system.

[0116] Figure 6 This is a schematic diagram of the structure of the agricultural knowledge recommendation device provided by the present invention. Figure 6 The agricultural knowledge recommendation device provided by the present invention is described. The agricultural knowledge recommendation device described below and the agricultural knowledge recommendation method provided by the present invention described above can be referred to each other. Figure 6 As shown, an agricultural knowledge recommendation device includes: a data acquisition module 601, a subspace determination module 602 and a knowledge recommendation module 603.

[0117] Data acquisition module 601, used to acquire basic attribute information and historical behavior data of target users;

[0118] A subspace determination module 602, configured to determine a time subspace, a region subspace, a preference subspace, and an interaction information subspace corresponding to the target user based on the basic attribute information and the historical behavior data;

[0119] The knowledge recommendation module 603 is used to obtain agricultural knowledge recommendation results for the target user from the first agricultural knowledge graph based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace;

[0120] Among them, the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0121] Among them, the time subspace, regional subspace, preference subspace and interactive information subspace are obtained based on basic attribute information and historical behavior data; the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0122] Specifically, the data acquisition module 601 , the subspace determination module 602 , and the knowledge recommendation module 603 may be electrically connected.

[0123] Optionally, the agricultural knowledge recommendation device may further include:

[0124] The simplification module is used to perform dimensionality reduction processing and multimodal information fusion on the second agricultural knowledge graph according to the semantics of the entity to obtain the first agricultural knowledge graph; wherein the second agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0125] Optionally, the knowledge recommendation module 603 can be specifically used to determine candidate paths in the first agricultural knowledge graph based on the time subspace, the geographical subspace, the preference subspace and the interactive information subspace; and obtain the agricultural knowledge recommendation results based on the candidate paths.

[0126] Optionally, the subspace determination module 602 can be specifically used to: obtain a time subspace matrix for characterizing the time subspace based on the current time information in the basic attribute information; obtain a regional subspace matrix for characterizing the regional subspace based on the regional information of the target user in the basic attribute information; obtain a preference subspace matrix for characterizing the preference subspace based on the crop rotation information, farming information, agricultural information and content information in the basic attribute information; obtain an interactive information subspace matrix for characterizing the interactive information subspace based on the number of clicks and browsing time of each knowledge by the target user in the historical behavior data.

[0127] Optionally, the agricultural knowledge recommendation device may further include: an evaluation module.

[0128] The evaluation module can be used to evaluate the agricultural knowledge recommendation result and obtain the evaluation score of the agricultural knowledge recommendation result.

[0129] The agricultural knowledge recommendation device provided in the embodiment of the present invention is used to execute the above-mentioned agricultural knowledge recommendation method of the present invention. Its implementation method is consistent with the implementation method of the agricultural knowledge recommendation method provided by the present invention and can achieve the same beneficial effects, which will not be repeated here.

[0130] The agricultural knowledge recommendation device is used in the agricultural knowledge recommendation method in the above embodiments. Therefore, the description and definition in the agricultural knowledge recommendation method in the above embodiments can be used for understanding each execution module in the embodiments of the present invention.

[0131] The agricultural knowledge recommendation device in the embodiment of the present invention determines the time subspace, regional subspace, preference subspace and interactive information subspace corresponding to the target user based on the basic attribute information and historical behavior data of the target user, and then comprehensively utilizes time and space, crops, varieties, target users and agricultural knowledge interaction information to perform knowledge reasoning on the first agricultural knowledge graph containing multimodal agricultural knowledge. It can effectively improve the comprehensive utilization rate of text, images, spatiotemporal information and human-computer interaction information in the agricultural intelligent knowledge service platform, ensure that agricultural knowledge reaches agricultural technicians and farmers quickly and accurately, support fast and accurate services, thereby improving production quality and efficiency, and with the continuous expansion and enrichment of data, realize more accurate and comprehensive knowledge hybrid recommendation, which can improve the accuracy, effectiveness and timeliness of recommending agricultural knowledge to agricultural science and technology personnel and farmers.

[0132] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communication interface 720 and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the agricultural knowledge recommendation method, which includes: obtaining basic attribute information and historical behavior data of the target user; based on the basic attribute information and historical behavior data, determining the time subspace, regional subspace, preference subspace and interactive information subspace corresponding to the target user; based on the time subspace, regional subspace, preference subspace and interactive information subspace, obtaining the agricultural knowledge recommendation result for the target user from the first agricultural knowledge graph; wherein the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0133] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0134] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the agricultural knowledge recommendation method provided by the above methods, and the method includes: obtaining basic attribute information and historical behavior data of the target user; based on the basic attribute information and historical behavior data, determining the time subspace, regional subspace, preference subspace and interactive information subspace corresponding to the target user; based on the time subspace, regional subspace, preference subspace and interactive information subspace, obtaining agricultural knowledge recommendation results for the target user from the first agricultural knowledge graph; wherein, the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0135] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the agricultural knowledge recommendation method provided by the above-mentioned methods, the method comprising: obtaining basic attribute information and historical behavior data of the target user; based on the basic attribute information and historical behavior data, determining the time subspace, geographical subspace, preference subspace and interactive information subspace corresponding to the target user; based on the time subspace, geographical subspace, preference subspace and interactive information subspace, obtaining agricultural knowledge recommendation results for the target user from a first agricultural knowledge graph; wherein the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge.

[0136] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0137] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for recommending agricultural knowledge, characterized in that: include: Obtain basic attribute information and historical behavior data of target users; Based on the basic attribute information and the historical behavior data, determining the time subspace, geographical subspace, preference subspace and interactive information subspace corresponding to the target user; Based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, an agricultural knowledge recommendation result for the target user is obtained from the first agricultural knowledge graph, and when performing multimodal agricultural knowledge reasoning based on the first agricultural knowledge graph, path encoding is used to define the initial point of the candidate path, and the self-attention operation of the residual connection is used to realize the addition of the path information, and the recommendation information is input into the embedding matrix to find the corresponding word embedding; Wherein, the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge; Before obtaining the agricultural knowledge recommendation result for the target user from the first agricultural knowledge graph based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, the method further includes: According to the semantics of the entity, the second agricultural knowledge graph is subjected to dimensionality reduction processing and multimodal information fusion to obtain the first agricultural knowledge graph; Wherein, the second agricultural knowledge graph is constructed based on multimodal agricultural knowledge; The multimodal information fusion of the second agricultural knowledge graph according to the semantics of the entity includes: Entities with the same semantics in the second agricultural knowledge graph are fused.

2. The agricultural knowledge recommendation method according to claim 1, characterized in that: The obtaining, from the first agricultural knowledge graph, an agricultural knowledge recommendation result for the target user based on the time subspace, the regional subspace, the preference subspace, and the interactive information subspace, comprises: Determine candidate paths in the first agricultural knowledge graph based on the time subspace, the geographical subspace, the preference subspace, and the interactive information subspace; Based on the candidate paths, the agricultural knowledge recommendation results are obtained.

3. The agricultural knowledge recommendation method according to claim 2, characterized in that: The determining, based on the basic attribute information and the historical behavior data, the time subspace, the region subspace, the preference subspace, and the interaction information subspace corresponding to the target user includes: Based on the current time information in the basic attribute information, obtaining a time subspace matrix for representing the time subspace; Based on the region information of the target user in the basic attribute information, obtaining a region subspace matrix for representing the region subspace; Based on the crop rotation information, farming information, agricultural condition information and content information in the basic attribute information, obtaining a preference subspace matrix for representing the preference subspace; Based on the number of clicks and browsing time of the target user on each piece of knowledge in the historical behavior data, an interactive information subspace matrix for characterizing the interactive information subspace is obtained.

4. The agricultural knowledge recommendation method according to claim 2, characterized in that: After obtaining the agricultural knowledge recommendation result for the target user from the first agricultural knowledge graph based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, the method further includes: The agricultural knowledge recommendation result is evaluated to obtain an evaluation score of the agricultural knowledge recommendation result.

5. An agricultural knowledge recommendation device, characterized in that: include: Data acquisition module, used to obtain basic attribute information and historical behavior data of target users; A subspace determination module, used to determine the time subspace, geographical subspace, preference subspace and interaction information subspace corresponding to the target user based on the basic attribute information and the historical behavior data; a knowledge recommendation module, for obtaining agricultural knowledge recommendation results for the target user from the first agricultural knowledge graph based on the time subspace, the regional subspace, the preference subspace and the interactive information subspace, and when performing multimodal agricultural knowledge reasoning based on the first agricultural knowledge graph, using path encoding to define the initial point of the candidate path, and using the self-attention operation of the residual connection to realize the addition of the path information, and inputting the recommended information into the embedding matrix to find the corresponding word embedding; Among them, the first agricultural knowledge graph is constructed based on multimodal agricultural knowledge; The knowledge recommendation module is further used to perform dimensionality reduction processing and multimodal information fusion on the second agricultural knowledge graph according to the semantics of the entity to obtain the first agricultural knowledge graph; Wherein, the second agricultural knowledge graph is constructed based on multimodal agricultural knowledge; The knowledge recommendation module performs multimodal information fusion on the second agricultural knowledge graph according to the semantics of the entity, including: Entities with the same semantics in the second agricultural knowledge graph are fused.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the agricultural knowledge recommendation method as described in any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the agricultural knowledge recommendation method according to any one of claims 1 to 4 is implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the agricultural knowledge recommendation method according to any one of claims 1 to 4 is implemented.

Images