Enterprise search methods, devices, equipment, computer storage media and programs

Abstract

This application discloses a method, apparatus, device, computer storage medium, and program for enterprise search. The method includes: determining an enterprise knowledge graph corresponding to a search task; determining a target retrieval index containing keywords from an index table of the enterprise knowledge graph, wherein the index table contains multiple retrieval indices, each containing keywords and the database location of the keyword index; inputting the target retrieval index into a pre-trained event search model, so that the event search model retrieves data related to the keywords from the database location of the keyword index according to the shortest path; and outputting the data related to the keywords as the search result corresponding to the search task. According to the embodiments of this application, an enterprise knowledge graph is constructed based on enterprise information, thereby realizing the organization and classification of enterprise information. Enterprise search based on the enterprise knowledge graph can improve the timeliness and accuracy of enterprise search, thus improving the efficiency of enterprise search.

Description

Technical Field

[0001] This application belongs to the field of enterprise information management technology, and in particular relates to an enterprise search method, apparatus, equipment, computer storage medium and program. Background Technology

[0002] Enterprise search refers to the use of search software to index various structured and unstructured information within an enterprise and provide retrieval methods. Within enterprises, senior decision-makers frequently use enterprise search systems to search for relevant information to assist in making various decisions and responding to unexpected events.

[0003] Because enterprises have a large amount of complex data, it is necessary to organize and classify enterprise information to ensure the timeliness and accuracy of information retrieval. Under such requirements, the enterprise search system needs to be significantly improved in terms of information organization and search speed. Otherwise, it will easily lead to a decrease in the overall efficiency of enterprise search and a decrease in the work efficiency of the enterprise search system. Therefore, there is an urgent need for an enterprise search method that can improve information organization capabilities and search speed. Summary of the Invention

[0004] This application provides a method, apparatus, device, computer storage medium, and program for enterprise search. It constructs an enterprise knowledge graph based on enterprise information, thereby realizing the organization and classification of enterprise information. Enterprise search based on the enterprise knowledge graph can improve the timeliness and accuracy of enterprise search, thus improving the efficiency of enterprise search.

[0005] In a first aspect, embodiments of this application provide a method for enterprise search, the method comprising:

[0006] Determine the enterprise knowledge graph corresponding to the search task. The search task contains keywords, and the enterprise knowledge graph is pre-built based on enterprise information.

[0007] The target retrieval index containing the keyword is determined from the index table of the enterprise knowledge graph. The index table contains multiple retrieval indexes, each of which contains the keyword and the database location of the keyword index. The database corresponds to the enterprise knowledge graph and stores enterprise data.

[0008] Input the target retrieval index into the pre-trained event search model so that the event search model can retrieve data related to the keywords from the database location of the keyword index according to the shortest path;

[0009] The retrieved data related to the keywords will be output as the search results for the search task.

[0010] Secondly, embodiments of this application provide an enterprise search device, the device comprising:

[0011] The knowledge graph determination module is used to determine the enterprise knowledge graph corresponding to the search task. The search task contains keywords, and the enterprise knowledge graph is pre-built based on enterprise information.

[0012] The index determination module is used to determine the target retrieval index containing the keywords from the index table of the enterprise knowledge graph. The index table contains multiple retrieval indexes, each of which contains the keywords and the database location of the keyword index. The database corresponds to the enterprise knowledge graph and stores enterprise data.

[0013] The data search module is used to input the target retrieval index into the pre-trained event search model, so that the event search model can retrieve data related to the keywords from the database location of the keyword index according to the shortest path;

[0014] The results output module is used to output the retrieved data related to the keywords as the search results for the search task.

[0015] Thirdly, embodiments of this application provide an electronic device, which includes: a processor and a memory storing computer program instructions;

[0016] When the processor executes the computer program instructions, it implements the steps of the enterprise search method as described in any embodiment of the first aspect.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the steps of the enterprise search method as described in any embodiment of the first aspect.

[0018] Fifthly, embodiments of this application provide a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to perform the steps of the enterprise search method as described in any embodiment of the first aspect.

[0019] The enterprise search method, apparatus, device, computer storage medium, and program of this application embodiment pre-construct an enterprise knowledge graph based on enterprise information. During enterprise search, the enterprise knowledge graph corresponding to the search task is determined, and a retrieval index containing keywords from the search task is determined from the search table of the enterprise knowledge graph. The retrieval index is input into an event search model, enabling the event search model to retrieve data corresponding to the keywords from the database corresponding to the enterprise knowledge graph according to the shortest path, and outputting the retrieved data as the search result. According to the embodiments of this application, constructing an enterprise knowledge graph based on enterprise information achieves the organization and classification of enterprise information. Enterprise search based on the enterprise knowledge graph can improve the timeliness and accuracy of enterprise search, thereby improving the efficiency of enterprise search. Data retrieval based on the retrieval index can reduce the search scope and improve search efficiency. Retrieving data based on the shortest path using the event search model can further shorten the retrieval time, thereby further improving search efficiency. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a business search method provided in an embodiment of this application;

[0022] Figure 2 This is a flowchart illustrating the process of determining a recommended data package in an enterprise search method provided in an embodiment of this application;

[0023] Figure 3 This is a flowchart illustrating another enterprise search method provided in an embodiment of this application;

[0024] Figure 4 This is a schematic diagram of the structure of an enterprise search device provided in an embodiment of this application;

[0025] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0027] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0028] Existing enterprise search systems are primarily based on client-side and server-side implementations. The client-side mainly includes a search scope management module, a data communication module, and a data processing module, while the server-side mainly includes a data communication module, a data storage and retrieval module, a data parsing module, and a search module. Client-side data is collected and assigned security attributes. Based on these attributes, protected index information is generated or the data itself is retrieved. This protected index information or the data itself is then uploaded to the server. The search scope management module tracks the data status of client devices and notifies the user of any changes in data status. Users can set the security attributes for this data. The server receives and saves the uploaded data, parses the client-side data and its own data, generates and saves ordinary index information, and when a search request is received, the server analyzes the request, performs the search, and returns the search results to the user.

[0029] Because enterprises have a large amount of complex data, it is necessary to organize and classify enterprise information to ensure the timeliness and accuracy of information retrieval. Under such requirements, the enterprise search system needs to be significantly improved in terms of information organization and search speed. Otherwise, it will easily lead to a decrease in the overall efficiency of enterprise search and a decrease in the work efficiency of the enterprise search system. Therefore, there is an urgent need for an enterprise search method that can improve information organization capabilities and search speed.

[0030] To address the problems of the prior art, embodiments of this application provide a method, apparatus, device, and computer storage medium for enterprise search. The enterprise search method provided in this application embodiment will be described first below.

[0031] Figure 1 A flowchart illustrating an embodiment of the enterprise search method provided in this application is shown. Figure 1 As shown, the method includes the following steps:

[0032] S110. Determine the enterprise knowledge graph corresponding to the search task. The search task contains keywords. The enterprise knowledge graph is pre-built based on enterprise information.

[0033] S120. Determine the target retrieval index containing the keywords in the search task from the index table of the enterprise knowledge graph. The index table contains multiple retrieval indexes, each of which contains keywords and the database location of the keyword index. The database corresponds to the enterprise knowledge graph and stores enterprise data.

[0034] S130. Input the target retrieval index into the pre-trained event search model so that the event search model can retrieve data related to the keyword from the database location of the keyword index according to the shortest path.

[0035] The database location of the keyword index refers to the database location of the keyword index contained in the target retrieval index.

[0036] S140. Output the retrieved data related to the keywords as the search results for the search task.

[0037] Therefore, according to the middleware configuration optimization method provided in this application embodiment, an enterprise knowledge graph is pre-constructed based on enterprise information. When performing an enterprise search, the enterprise knowledge graph corresponding to the search task is determined, and a retrieval index containing keywords from the search task is determined from the search table of the enterprise knowledge graph. The retrieval index is input into the event search model, so that the event search model retrieves the data corresponding to the keywords from the database corresponding to the enterprise knowledge graph according to the shortest path, and outputs the retrieved data as the search result. According to this application embodiment, the construction of an enterprise knowledge graph based on enterprise information realizes the organization and classification of enterprise information. Enterprise search based on the enterprise knowledge graph can improve the timeliness and accuracy of enterprise search, thereby improving the efficiency of enterprise search. Data retrieval based on the retrieval index can reduce the search scope and improve search efficiency. Retrieving based on the shortest path through the event search model can further shorten the time required for retrieval, thereby further improving search efficiency.

[0038] In some embodiments, an enterprise knowledge graph may include multiple historical versions of the enterprise knowledge graph, and different historical versions of the enterprise knowledge graph can be used to complete different search tasks. Enterprise information may include conventional numerical information, text information, image information, video information, and voice information, etc.

[0039] As an example, pre-constructing an enterprise knowledge graph based on enterprise information can begin by first acquiring multimodal information and data related to the enterprise through interfaces or web crawlers. This multimodal information can include conventional numerical information, text information, image information, video information, and voice information. After acquiring the enterprise information, it can be preprocessed using data cleaning and other preprocessing methods, and a reasonable classification and storage method can be selected according to the format of the enterprise information. Then, feature extraction can be performed on the enterprise information. Specifically, large-scale data can be used, combined with statistical learning and logical rule methods, to train the multimodal data feature extraction of enterprise information. Based on the text features, image features, video features, and voice features corresponding to the aforementioned text information, image features, video features, and voice information, neural network models for text feature extraction, image feature extraction, video feature extraction, and voice feature extraction can be developed respectively to obtain text data features, image data features, video data features, and voice data features. Conventional numerical data features can also be obtained through logical rule methods.

[0040] Furthermore, to obtain the representational features corresponding to each data feature, a unified representation, association analysis, commonality selection, and coarse classification are performed based on the obtained data features. For example, deep learning technology is first used for unified representation, and then clustering, association, and distance algorithms are used to perform association analysis and classification on each data feature. After classification, multimodal mixed data features of each category can be obtained. Thus, through statistical learning methods, logical rules, and deep learning methods, the common features of each category in the coarse representation, i.e., representational features, can be obtained. After obtaining the representational features corresponding to each data feature, the representational features can be stored in a pre-defined database, and then a NoSQL (Not Only SQL) database is used. NoSQL's key-value storage method is used for the storage and management of multimodal data. Through the feature extraction model of multimodal data of enterprise information, the representational features of each category are obtained as indexes, and text data, image data, video data, and voice data in the category are used as corresponding values, which are then stored in the database. Feature information can be used as an index to effectively manage and use data.

[0041] To establish the foundation for a knowledge graph, entities, attributes, and relationships can be extracted from multimodal data. These categories serve as the basis for constructing the knowledge graph. Entity extraction can involve extracting atomic information from text, such as names, organization names, locations, times, and amounts of money. Relationship extraction can be based on the relationships between extracted entities. Statistical learning methods, logical rule methods, and deep learning methods can be used to perform conventional entity extraction, attribute extraction, and relationship extraction analysis on image, audio, text, and video data in each category to obtain entity, attribute, and relationship information. This effectively reduces the workload of constructing the knowledge graph. Further, association analysis and cross-validation are performed between enterprise information, and cross-association analysis is conducted across enterprise information across categories to obtain new entities, attributes, and relationships. This cross-modal association analysis and cross-validation remove erroneous information. Using the obtained high-confidence entity, attribute, and relationship information as the foundation for constructing the knowledge graph enhances the credibility of its core elements.

[0042] Finally, based on the acquired knowledge graph, a unified representation of enterprise knowledge graphs across modal data is constructed. Here, conventional representation methods can be used to uniformly represent the enterprise knowledge graph. Based on conventional knowledge graph construction techniques, knowledge reasoning research is conducted on the constructed enterprise knowledge graph to establish hidden relationship graphs between entities, thereby obtaining the expanded enterprise knowledge graph.

[0043] Thus, based on multiple pre-built historical versions of enterprise knowledge graphs, the corresponding enterprise knowledge graph for a search task can be determined. Different search tasks can be completed based on different historical versions of the knowledge graph. For example, before the search, multiple search tasks corresponding to enterprise knowledge graphs can be scheduled in the system using a pre-built mathematical model, thereby improving efficiency. Furthermore, the database records historical information on enterprise searches performed using various historical versions of enterprise knowledge graphs for different types of enterprise information. This historical information includes the processing time taken by each historical version of the enterprise knowledge graph to complete the corresponding enterprise search for each type of enterprise information.

[0044] Based on this, as an example, in S110, the determination of the enterprise knowledge graph corresponding to the search task can specifically include:

[0045] Pre-build n versions of the enterprise knowledge graph. When n search tasks are received, determine the enterprise knowledge graph version corresponding to each of the n search tasks according to the pre-built mathematical model.

[0046] For each search task, the corresponding version of the enterprise knowledge graph is used as the enterprise knowledge graph for that search task.

[0047] The mathematical models are shown in formulas (1), (2), and (3) below:

[0048]

[0049]

[0050]

[0051] Where minz is the objective function, representing the minimization of the total resources consumed in completing all n search tasks, x ij c represents the number of resources consumed by the i-th version of the enterprise knowledge graph to complete the j-th search task. ij Let s·t be a constant term, and let s·t represent the constraint condition, where This indicates that the i-th version of the enterprise knowledge graph is only responsible for one search task. This indicates that the j-th search task can only be handled by one version of the enterprise knowledge graph, x ij =0 or 1 represents x ij It can only take the value 0 or 1.

[0052] Furthermore, based on the aforementioned mathematical model, the solution can be obtained using, for example, the matrix covering method or by directly employing a 0-1 programming approach in MATLAB. The matrix covering method can include the following steps:

[0053] Step 1: Find the equivalence distribution matrix (subtract the smallest element from each row and column);

[0054] Step 2: Find the independent zero elements and add a marker box (zeros that are not in the same column or row);

[0055] Step 3: The optimal decision is to stop the calculation when n independent zero elements are reached;

[0056] Step 4: Find the coverage line: Block rows without zero-element markers and mark them with a checkmark; also block columns without zero-element markers in the blocked rows, and also block rows with zero-element markers in the blocked columns; draw a coverage line between the unblocked rows and the blocked columns.

[0057] Step 5: Adjust the allocation matrix: Select the smallest element k in the uncovered elements, subtract k from the uncovered rows, add k to the covered columns, and go to step 2.

[0058] Therefore, when multiple search tasks are received, the enterprise knowledge graph corresponding to the search task can be determined from multiple historical versions of the enterprise knowledge graph through a pre-built mathematical model, which greatly improves search efficiency.

[0059] In some embodiments, after constructing the enterprise knowledge graph, an index table can be added to each enterprise knowledge graph to improve search efficiency. Thus, in S120, the retrieval index corresponding to the keywords contained in the search task, i.e., the target retrieval index, can be determined from the index table.

[0060] The index table contains multiple search indexes, each containing keywords and the database location of the keyword index. The database corresponds to the enterprise knowledge graph and stores enterprise data.

[0061] The retrieval index in the index table can be an inverted index. An inverted index is a data structure that represents a mapping, such as indexing by a word, phrase, or number as a key, mapping to all documents or database files containing that word or phrase. The database corresponding to the enterprise knowledge graph stores enterprise data, such as enterprise data corresponding to enterprise information.

[0062] As an example, the index table of an enterprise knowledge graph uses an inverted index, which consists of three parts: a term index, a term dictionary, and a posting list. Thus, after determining the target retrieval index containing keywords in the enterprise knowledge graph's index table, the database location of the keyword index can be determined based on the term index in the term dictionary, which is the database corresponding to the enterprise knowledge graph. The term index can have various dictionary structures, such as hash tables, B-trees, B+ trees, and FSTs.

[0063] In some embodiments, in S130, the shortest path can be the shortest path between nodes in the enterprise knowledge graph, for example, determined by identifying the distance between the nodes in the enterprise knowledge graph. Each node can include basic elements such as events, locations, and participants, and can be a graph-structured knowledge fragment triggered by an action or a change in state. Therefore, the event search model can retrieve data related to keywords from the keyword-indexed database location based on the shortest path.

[0064] As an example, in order to retrieve data related to keywords based on the shortest path using an event search model, an event search model can be constructed before step S130 above. Constructing the event search model may include:

[0065] Perform knowledge retrieval on the enterprise knowledge graph to determine the shortest path between each node;

[0066] An event search model is constructed based on the shortest path between nodes and deep learning algorithms.

[0067] Deep learning algorithms can include, but are not limited to, the following methods:

[0068] Neural network algorithms: A neural network is a computational system with interconnected nodes that function much like neurons in the human brain. These neurons process and transmit information between each other. Each neural network is a series of algorithms that attempt to identify potential relationships in a set of data by simulating the process of the human brain.

[0069] Backpropagation is a very popular supervised learning algorithm used to train feedforward neural networks.

[0070] Feedforward neural network algorithms are typically fully connected, meaning that each neuron in a layer is connected to all other neurons in the next layer.

[0071] In addition to aiding the vision of robots and autonomous vehicles, convolutional neural network algorithms have been successfully applied in fields such as facial recognition, object detection, and traffic sign recognition.

[0072] Recurrent neural network algorithms have been very successful in many NLP (Natural Language Processing) tasks. In traditional neural networks, it can be understood that all inputs and outputs are independent.

[0073] Recurrent neural network algorithms are another form of recurrent network, the difference being that they are tree-like structures. Therefore, they can model hierarchical structures in the training dataset.

[0074] Autoencoders can recover the input signal at the output. They have a hidden layer inside. Autoencoders are designed not to accurately copy the input to the output, but in order to minimize the error, the network is forced to learn to select the most important features.

[0075] The Restricted Boltzmann Machine (RBM) algorithm is a stochastic neural network (a neural network means that we have neuron-like units whose binary activation depends on the neighboring units they are connected to).

[0076] GAN (Generative Adversarial Networks) algorithms are becoming a popular machine learning model for online retail because they are able to understand and reconstruct visual content with increasingly higher accuracy.

[0077] Graph neural network algorithms aim to model graph data, meaning they identify and numerically represent the relationships between nodes in the graph. These can then be used in various other machine learning models for tasks such as clustering and classification.

[0078] As an example, taking the construction of an event search model based on graph neural network algorithms, knowledge retrieval is performed on an enterprise knowledge graph. Then, a graph neural network algorithm can be used to model the shortest path for event retrieval based on the corresponding graph data. This can identify the relationships between nodes in the enterprise knowledge graph and represent them numerically. The graph neural network here typically consists of two modules: a propagation module and an output module.

[0079] The propagation module is used to transmit information and update the state between nodes in the enterprise knowledge graph; the aggregator is used to learn the latent representation h of a node v by aggregating the information of its surrounding nodes based on the following formula (4). v (state embedding). The Updater is used to update the state embedding of a node based on the following formula (5).

[0080] h v =f(X) v ,X co[v] ,h ne[v] ,X ne[v] (4)

[0081] H t+1 =F(H t ,X) (5)

[0082] Among them, X v For the feature information of node v, X co[v] For the features of its surrounding edges, h ne[v] X represents the state embedding of the neighboring nodes around node v. ne[v] Represents the features of surrounding nodes.

[0083] The output module can define the objective function based on the vector representation of nodes and edges according to different tasks, as shown in the following formula (6).

[0084]

[0085] In supervised learning scenarios, for a specific node, its supervision signal is represented as: t v The loss function is defined as shown in formula (7).

[0086]

[0087] Therefore, by training a model of the shortest path between nodes in the enterprise knowledge graph using deep learning algorithms, and based on the obtained event search model, the efficiency of retrieving data related to keywords from the database location indexed by keywords is improved.

[0088] In some embodiments, before outputting search results, in order to improve the intelligence and efficiency of enterprise search, enterprise data can be packaged into enterprise data packets. Accordingly, the specific implementation of S140 above may include:

[0089] The retrieved data related to the keywords is packaged into enterprise data packages to obtain the enterprise data packages related to the keywords; the enterprise data packages related to the keywords are then output as the search results corresponding to the search task.

[0090] Each enterprise data package has different content, but all contain several data feature values. The enterprise data packages are presented in a conventional format, such as forms, workflows, or statistics. Thus, the feature content of the enterprise data packages can be extracted, with the data feature values ​​representing the proportion of that feature content within the total content.

[0091] Based on this, the data related to the keywords is output in the form of enterprise data packages, so that users can obtain the enterprise data packages corresponding to the search task.

[0092] Furthermore, such as Figure 2 As shown, in some embodiments, the enterprise search method may further include:

[0093] S210. Obtain the user's rating of the enterprise data package they have used. The rating refers to the user's preference score for the enterprise data package.

[0094] S220. Based on the scoring, construct a scoring matrix for enterprise data packets;

[0095] S230. Based on the scoring matrix and the enterprise data package related to the keywords, determine the recommended data package, wherein the enterprise data package related to the keywords is the enterprise data package obtained in S140.

[0096] S240. Output the recommended data package as well as the search results corresponding to the search task.

[0097] The rating is obtained by users rating the enterprise data package after using it, but users can also choose not to rate it. Therefore, the rating matrix of the enterprise data package is sparse when it is constructed based on the rating.

[0098] As an example, due to the sparsity of the rating matrix, it is necessary to complete the sparse rating matrix before determining the recommended data package based on preference. The rating matrix R can be approximated by the product of two matrices, as shown in the following formula (8):

[0099] R≈P T Q(8)

[0100] The rating matrix R is m*n, containing m users and n enterprise data packets. Each user cannot use all enterprise data packets simultaneously; only a portion of the enterprise data packets are rated. P is k*m dimensional, with the i-th column vector serving as the feature p of user i. i This feature p i Q is k-dimensional; Q is k*n-dimensional, and the j-th column vector is taken as the feature q of enterprise data packet j. j This feature q j It is also k-dimensional.

[0101] For example, given 100 users and 1000 enterprise data packages, users rate the enterprise data packages they have used, thus forming a user-enterprise data package rating matrix. To distinguish between different users and enterprise data packages, three types of features are given, k=3, namely form type, process type, and statistical type; the feature of the first user is p1=[0.8,0.2,0.1]. T This means that users prefer form-based content. The characteristic of the second enterprise data package is q2 = [0.3, 0.2, 0.6]. T This means that this enterprise data package is more statistically oriented, so the predicted rating for the first user after using the second enterprise data package is 11. T q2 = 0.34, yielding a comprehensive level of liking.

[0102] Therefore, to obtain the above P and Q, and make P T Q is close to R, and there exists a loss function minL as shown in the following formula (9).

[0103]

[0104] Further differentiation of the loss function minL yields the following formulas (10) and (11).

[0105]

[0106]

[0107] Where λ is the regularization coefficient, which needs to be tuned, and I is the identity matrix. The update strategy can be obtained by satisfying the following formulas (12) and (13).

[0108]

[0109]

[0110] Where α represents the learning rate, and finally, through the above iterative updates, P and Q are obtained, thus yielding an approximate scoring matrix P. T Q, thus completing the rating matrix R.

[0111] Therefore, recommended data packages can be determined based on the completed scoring matrix R. Based on the preference type of enterprise data packages related to keywords obtained through direct search, other enterprise data packages of the same preference type can be automatically recommended as recommended data packages. By determining the recommended data packages, efficient and intelligent enterprise search capabilities can be achieved.

[0112] As an example, to conduct enterprise searches more intelligently and efficiently, the aforementioned S230 may specifically include:

[0113] Determine the preferred type of enterprise data packages related to keywords;

[0114] Based on preference type and the rating matrix, enterprise data packages whose ratings meet preset conditions are selected as recommended data packages.

[0115] As an example, based on the preference type of enterprise data packages related to keywords, recommended data packages can be selected based on the rating matrix according to the preference type. The method for determining the recommended data packages can be either based on the completed rating matrix R, selecting the top five highest-rated enterprise data packages in the preference type obtained from direct search; or based on the completed rating matrix R, selecting the highest-rated enterprise data packages in the preference type obtained from direct search, and then selecting the ones with the highest current user preference (i.e., the highest user rating).

[0116] Therefore, by using enterprise knowledge graph-based enterprise search, different business personnel can quickly and efficiently perform enterprise searches, add search content, classify and intelligently search / recommend, making enterprise search more intelligent and efficient.

[0117] Enterprise information and data are usually updated frequently. In order to ensure that the enterprise knowledge graph can still be used after the enterprise information and data are updated, the updated data will be added to the enterprise knowledge graph as additional content to realize the updating of the enterprise knowledge graph.

[0118] To add content to the enterprise knowledge graph, this application also provides another embodiment of the enterprise search method. For example... Figure 3 As shown, the method may further include:

[0119] S310. When it is necessary to add content to the enterprise knowledge graph, the content to be added shall be classified and the category to which the content to be added belongs shall be determined.

[0120] S320. Add the content to be added to the category to which the content to be added belongs in the enterprise knowledge graph.

[0121] In some embodiments, the category to which the content to be added belongs can be determined, for example, based on the distance between the content to be added and various category information points. In this way, the content to be added can be added to the category corresponding to the content in the enterprise knowledge graph.

[0122] As an example, the aforementioned enterprise knowledge graph contains multiple categories of content. To determine the category to which the content to be added belongs, S310 may specifically include:

[0123] Determine the distance between the content to be added and each category in the enterprise knowledge graph;

[0124] Select the category with the shortest distance to the content to be added from among multiple categories as the category to which the content to be added belongs.

[0125] As an example, when it is necessary to add content to different categories of content in an enterprise knowledge graph, the content to be added can be classified according to the following formula (14).

[0126]

[0127] Where, d i This represents the distance between the point to be added and the feature information points of each category. These points are points with coordinates in a two-dimensional coordinate system. This system uses the two feature information data values ​​of all categories within the enterprise knowledge graph as the horizontal and vertical axes. The content to be added also contains these two feature information data values, and the coordinates of symbolic content for each category are pre-displayed within this two-dimensional coordinate system. Here, x1 represents the horizontal coordinate of the point to be added, and y1 represents the horizontal coordinate of the point. i x1 represents the ordinate of the point where content to be added, x2 represents the x-coordinate of the symbolic content points in each category, and y1 represents the coordinates of the points where content to be added is added. i The ordinate of the coordinate point representing the symbolic content of each category.

[0128] To improve the accuracy of classification and the efficiency of comparison distance, the adjustment function s(d) of the following formulas (15), (16) and (17) can be introduced. i ).

[0129] d s =s(d i )*d i (15)

[0130]

[0131] f(d i )=[d i ]*(1+(d i -[d i (17)

[0132] Where, d s d represents the distance after processing by the adjustment function. k The distance from which data is discarded can be set and adjusted by business personnel or managers, which is beneficial for pre-emptively removing classification choices that are far apart; f(d i ) represents the distance transformation function, used to achieve adaptive augmentation according to the weights and the cardinality of the distance itself, which is beneficial for revealing the differences between various distances; [d i ] indicates that d i Round down.

[0133] Therefore, the above d can be... s The minimum value in the table corresponds to the category as the classification target, and the content to be added is classified into the category to which the coordinate point of the corresponding symbolic content belongs, so as to quickly add content and automatically classify it, thereby improving the system's organization efficiency.

[0134] It should be noted that the application scenarios described in the above embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0135] Based on the same inventive concept, this application also provides a business search device. (Specifically combined with...) Figure 4 Please provide a detailed explanation.

[0136] Figure 4 This is a schematic diagram of the structure of an enterprise search device provided in an embodiment of this application.

[0137] like Figure 4 As shown, the enterprise search device 400 may include:

[0138] The graph determination module 401 is used to determine the enterprise knowledge graph corresponding to the search task. The search task contains keywords, and the enterprise knowledge graph is pre-built based on enterprise information.

[0139] The index determination module 402 is used to determine the target retrieval index containing the keyword from the index table of the enterprise knowledge graph. The index table contains multiple retrieval indexes, each of which contains the keyword and the database location of the keyword index. The database corresponds to the enterprise knowledge graph and stores enterprise data in the database.

[0140] The data search module 403 is used to input the target retrieval index into the pre-trained event search model so that the event search model can retrieve data related to the keyword from the database location of the keyword index according to the shortest path;

[0141] The result output module 404 is used to output the retrieved data related to the keywords as the search results corresponding to the search task.

[0142] In some embodiments, the above-described map determination module 401 can be specifically used for:

[0143] Pre-build n versions of the enterprise knowledge graph. When n search tasks are received, determine the enterprise knowledge graph version corresponding to each of the n search tasks according to the pre-built mathematical model.

[0144] For each search task, the corresponding version of the enterprise knowledge graph is used as the enterprise knowledge graph for the search task.

[0145] The mathematical model is shown below:

[0146]

[0147]

[0148]

[0149] In the formula, minz is the objective function, representing the minimization of the total resources consumed to complete all n search tasks, and x ij c represents the number of resources consumed by the i-th version of the enterprise knowledge graph to complete the j-th search task. ij Let s·t be a constant term, and let s·t represent the constraint condition, where This indicates that the i-th version of the enterprise knowledge graph is only responsible for one search task. This indicates that the j-th search task can only be handled by one version of the enterprise knowledge graph, x ij =0 or 1 represents x ij It can only take the value 0 or 1.

[0150] In some embodiments, in order to retrieve keyword-related data based on the shortest path using an event search model, the enterprise search device 400 may further include:

[0151] The first determination module is used to perform knowledge retrieval on the enterprise knowledge graph and determine the shortest path between each node.

[0152] The training module is used to build an event search model based on the shortest path between nodes and deep learning algorithms.

[0153] In some embodiments, the above-mentioned result output module 404 is specifically used for:

[0154] The retrieved data related to the keyword is packaged into enterprise data packages to obtain enterprise data packages related to the keyword;

[0155] The data packages of enterprises related to the keywords will be output as the search results for the search task.

[0156] In some embodiments, the above-mentioned enterprise search device 400 may further include:

[0157] The acquisition module is used to obtain users' ratings of the enterprise data packages they have used. The rating refers to the user's preference score for the enterprise data package.

[0158] The building module is used to construct a scoring matrix for enterprise data packages based on the scores.

[0159] The second determination module is used to determine the recommended data package based on the scoring matrix and the enterprise data package related to the keywords;

[0160] The output module is used to output the recommended data package as well as the search results corresponding to the search task.

[0161] In some embodiments, to perform enterprise searches more intelligently and efficiently, the second determining module may specifically include:

[0162] The first page determines the submodule, which is used to determine the preferred type of enterprise data packages related to keywords;

[0163] The first selection submodule is used to select enterprise data packages whose ratings meet preset conditions based on the preference type and the rating matrix, as recommended data packages.

[0164] In some embodiments, the above-mentioned enterprise search device 400 may further include:

[0165] The third module is used to categorize the content to be added to the enterprise knowledge graph and determine the category to which the content to be added belongs.

[0166] The Add module is used to add content to the category to which the content to be added belongs in the enterprise knowledge graph.

[0167] In some embodiments, the enterprise knowledge graph described above contains multiple categories of content. To determine the category to which the content to be added belongs, the third determining module may specifically include:

[0168] The second determination submodule is used to determine the distance between the content to be added and each category in the enterprise knowledge graph;

[0169] The second selection submodule is used to select the category with the shortest distance to the content to be added from among multiple categories as the category to which the content to be added belongs.

[0170] Therefore, according to the middleware configuration optimization method provided in this application embodiment, an enterprise knowledge graph is pre-constructed based on enterprise information. When performing an enterprise search, the enterprise knowledge graph corresponding to the search task is determined, and a retrieval index containing keywords from the search table of the enterprise knowledge graph is determined. The retrieval index is input into the event search model, so that the event search model retrieves the data corresponding to the keywords from the database corresponding to the enterprise knowledge graph based on the shortest path, and outputs the retrieved data as the search result. According to this application embodiment, constructing an enterprise knowledge graph based on enterprise information realizes the organization and classification of enterprise information. Enterprise search based on the enterprise knowledge graph can improve the timeliness and accuracy of enterprise search, thereby improving the efficiency of enterprise search. Data retrieval based on the retrieval index can reduce the search scope and improve search efficiency. Retrieving based on the shortest path through the event search model can further shorten the time required for retrieval, thereby further improving search efficiency.

[0171] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application is shown.

[0172] The electronic device 500 may include a processor 501 and a memory 502 storing computer program instructions.

[0173] Specifically, the processor 501 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0174] Memory 502 may include a large-capacity memory for data or instructions. For example, and not limitingly, memory 502 may include a hard disk drive (HDD), a floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 502 may include removable or non-removable (or fixed) media. Where appropriate, memory 502 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 502 is non-volatile solid-state memory. Memory 502 may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, typically, memory 502 includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it performs the operations described in any of the enterprise search methods in the above embodiments.

[0175] The processor 501 implements any of the enterprise search methods described in the above embodiments by reading and executing computer program instructions stored in the memory 502.

[0176] In one example, electronic device 500 may also include communication interface 505 and bus 510. Wherein, as... Figure 5 As shown, the processor 501, memory 502, and communication interface 505 are connected through bus 510 and complete communication with each other.

[0177] The communication interface 505 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0178] Bus 510 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 510 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0179] For example, the electronic device 500 can be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc.

[0180] The electronic device 500 can execute the enterprise search method in the embodiments of this application, thereby achieving a combination Figure 1 and Figure 4 Described enterprise search methods and apparatus.

[0181] Furthermore, in conjunction with the enterprise search methods described in the above embodiments, this application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the enterprise search methods described in the above embodiments.

[0182] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0183] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0184] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0185] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0186] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for enterprise search, characterized in that, include: The enterprise knowledge graph corresponding to the search task is determined. The search task contains keywords. The enterprise knowledge graph is pre-constructed based on enterprise information. Pre-constructing the enterprise knowledge graph based on enterprise information includes: performing association analysis and classification on enterprise information to obtain the representation features and multimodal data of each category; obtaining entities, attributes and relationships based on the multimodal data in each category; and constructing the enterprise knowledge graph based on entities, attributes and relationships. The target retrieval index containing the keyword is determined from the index table of the enterprise knowledge graph. The index table contains multiple retrieval indexes, each of which contains the keyword and the database location of the keyword index. The database corresponds to the enterprise knowledge graph and stores enterprise data, which includes the representation features of each category and multimodal data. The database uses the representation features of each category as the database location and the multimodal data in the category as the value corresponding to the database location. The target retrieval index is input into a pre-trained event search model so that the event search model retrieves data related to the keyword from the database location of the keyword index according to the shortest path; The retrieved data related to the keywords will be output as the search results for the search task. The determination of the enterprise knowledge graph corresponding to the search task includes: Pre-build n versions of the enterprise knowledge graph. When n search tasks are received, determine the enterprise knowledge graph version corresponding to each of the n search tasks according to the pre-built mathematical model. For each search task, the corresponding enterprise knowledge graph version is used as the enterprise knowledge graph for that search task, so as to minimize the total resources consumed in completing all n search tasks.

2. The method according to claim 1, characterized in that, The mathematical model is shown below: In the formula, Let be the objective function, representing the goal of minimizing the total resources consumed in completing all n search tasks. Indicates the first The first version of the enterprise knowledge graph has been completed. The number of resources consumed by each search task. For constant terms, Describes the constraint conditions, where, Indicates the first Each version of the enterprise knowledge graph is only responsible for one search task. Indicates the first Each search task can only be handled by one version of the enterprise knowledge graph. express It can only take the value 0 or 1.

3. The method according to claim 1, characterized in that, Before inputting the target retrieval index into the pre-trained event search model, the method further includes: Perform knowledge retrieval on the enterprise knowledge graph to determine the shortest path between each node; An event search model is constructed based on the shortest path between the nodes and a deep learning algorithm.

4. The method according to claim 1, characterized in that, The step of outputting the retrieved data related to the keyword as the search results corresponding to the search task includes: The retrieved data related to the keyword is packaged into enterprise data packages to obtain enterprise data packages related to the keyword; The data packages of enterprises related to the keywords will be output as the search results for the search task.

5. The method according to claim 4, characterized in that, The method further includes: Obtain user ratings for enterprise data packages they have used, where the ratings refer to the user's preference score for the enterprise data packages; Based on the scores, a scoring matrix for the enterprise data packets is constructed; Based on the scoring matrix and the enterprise data package related to the keywords, a recommended data package is determined; The recommended data package is also output as the search result corresponding to the search task.

6. The method according to claim 5, characterized in that, The step of determining the recommended data package based on the scoring matrix and the enterprise data package related to the keywords includes: Determine the preference type of enterprise data packets related to the keywords; Based on the preference type and the rating matrix, enterprise data packages whose ratings meet preset conditions are selected as recommended data packages.

7. The method according to claim 1, characterized in that, The method further includes: When it is necessary to add content to the enterprise knowledge graph, the content to be added is categorized to determine the category to which the content to be added belongs; Add the content to be added to the category to which the content to be added belongs in the enterprise knowledge graph.

8. The method according to claim 7, characterized in that, The enterprise knowledge graph contains multiple categories of content. The process of categorizing the content to be added and determining its category includes: Determine the distance between the content to be added and each category in the enterprise knowledge graph; The category with the shortest distance to the content to be added among the multiple categories is selected as the category to which the content to be added belongs.

9. A business search device, characterized in that, include: The graph determination module is used to determine the enterprise knowledge graph corresponding to the search task, wherein the search task contains keywords and the enterprise knowledge graph is pre-built based on enterprise information. The pre-construction of the enterprise knowledge graph based on enterprise information includes: performing association analysis and classification on enterprise information to obtain the representational features and multimodal data of each category; obtaining entities, attributes, and relationships based on the multimodal data in each category; and constructing the enterprise knowledge graph based on the entities, attributes, and relationships. The step of determining the enterprise knowledge graph corresponding to a search task includes: pre-constructing n versions of the enterprise knowledge graph; upon receiving n search tasks, determining the enterprise knowledge graph version corresponding to each of the n search tasks according to a pre-constructed mathematical model; and for each search task, using the enterprise knowledge graph of its corresponding version as the enterprise knowledge graph corresponding to the search task, so as to minimize the total resources consumed in completing all n search tasks. An index determination module is used to determine a target retrieval index containing the keyword from the index table of the enterprise knowledge graph. The index table contains multiple retrieval indexes, each of which contains the keyword and the database location of the keyword index. The database corresponds to the enterprise knowledge graph and stores enterprise data. The enterprise data includes the representational features of each category and multimodal data. The database uses the representational features of each category as the database location and the multimodal data in the category as the value corresponding to the database location. The data search module is used to input the target retrieval index into a pre-trained event search model, so that the event search model can retrieve data related to the keyword from the database location of the keyword index according to the shortest path; The result output module is used to output data related to the keyword as the search results corresponding to the search task.

10. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the enterprise search method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the enterprise search method as described in any one of claims 1-8.

12. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device causes the electronic device to perform the enterprise search method as described in any one of claims 1-8.

Images