Method, device, computer device and readable medium for recommending a bearer network solution

Abstract

The present disclosure provides a kind of bearer network scheme recommendation method, concept knowledge graph for indicating the relationship between concept and instance knowledge graph for indicating the relationship between instance are pre-set in the bearer network recommendation device, and instance is the instance of each concept in concept knowledge graph;Method includes: determining scheme type according to received recommendation request;According to the mapping relationship of pre-set scheme type and meta path set, the meta path set corresponding to the scheme type is determined, the meta path set includes multiple meta paths, and the meta path indicates the path between the user request and scheme in the concept knowledge graph;According to the scheme type, determine scheme template;For each sub-scheme in the scheme template, according to the recommendation request, the meta path set and the instance graph, the recommendation probability of alternative scheme is calculated;At least according to the recommendation probability of the alternative scheme in each sub-scheme and the scheme template, generate recommended scheme.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of communication, in particular to a bearer network scheme recommendation method and device, computer equipment and readable medium. BACKGROUND

[0002] The bearer network integrates SDH (Synchronous Digital Hierarchy) / MSTP (Multi-Service Transfer Platform), PTN (Packet Transport Network), IPRAN (IP Radio Access Network) and WDM (Wavelength Division Multiplexing) / OTN (Optical Transport network) transmission technologies, and can be logically divided into four levels: access layer, convergence layer, core layer and backbone layer. SDH / MSTP is an earlier transmission technology, which flourished from the 1990s to the beginning of this century. The characteristics of SDH are block frame structure, rich OAM overhead, flexible service scheduling and perfect protection function. PTN technology takes packet as the transmission unit, and its frame structure is no longer a standard block structure, but can vary in length. PTN technology can better carry telecom-grade Ethernet services, and meet the five basic attributes of standardization, high reliability, flexible scalability, strict QoS and perfect OAM (Operation Administration and maintenance). IPRAN is an IP-based wireless access network, which is produced to meet the needs of X2 interface and S1-FLEX services in the LTE (Long Term Evolution) stage. Compared with the PTN and router joint networking scheme, IPRAN can save the investment of routers, realize global optimization of network resources, and also rely on the advantages of IP-based packet transmission to provide more flexible and reliable network than PTN. The characteristics of WDM are large transmission capacity, so it is usually used in backbone networks, and is also widely used in metropolitan area networks in large and medium-sized cities. However, compared with SDH, it has some fatal defects, such as lack of OAM, inflexible scheduling, and imperfect protection. In order to make up for these defects, OTN was born.

[0003] There are various types of solutions for transport networks, including deployment solutions, maintenance solutions, and cutover solutions. For example, the deployment solution for IPRAN is a deployment solution, while the automatic scaling solution for VPN (Virtual Private Network) is a maintenance solution. A solution consists of one or more sub-solutions, and each sub-solution may also consist of a series of sub-solutions. The relationship between the solution and its sub-solutions is as follows: Figure 1 As shown, a certain type of solution may include: sub-solution A, sub-solution B, and sub-solution C. Sub-solution B includes sub-solutions D and E. Taking a deployment solution as an example, the sub-solutions of a deployment solution may include: business model solution, reliability solution, clock synchronization deployment solution, and management plane solution. Among them, the sub-solutions of the business model solution may include: VPN solution, tunnel solution, etc. A solution may include one or more strategies; for example, a deployment solution may include a resource selection strategy.

[0004] In the past, the selection and formulation of transport network solutions relied on the experience of human experts, and different people might have different judgments on the application of the same solution. Moreover, due to factors such as incomplete documentation and personnel turnover, the transmission of expert experience is uncertain, and there is a possibility that the experience gained in solution formulation may be lost due to personnel turnover. Transport network solutions rely entirely on the experience of human experts, making it impossible to deploy and maintain the network directly according to user intentions. Summary of the Invention

[0005] This disclosure addresses the aforementioned deficiencies in the prior art by providing a method, apparatus, computer equipment, and readable medium for recommending bearer network solutions.

[0006] In a first aspect, embodiments of this disclosure provide a method for recommending bearer network schemes. The method is applied to a bearer network scheme recommendation device, which has a pre-set concept knowledge graph representing relationships between concepts and an instance knowledge graph representing relationships between instances. The concepts include: user requests, scheme selection factors, and schemes. The instances are instances of each concept in the concept knowledge graph. The method includes:

[0007] Determine the solution type based on the received recommendation request;

[0008] The meta-path set corresponding to the scheme type is determined according to the preset mapping relationship between scheme type and meta-path set. The meta-path set includes multiple meta-paths, and the meta-paths represent the paths from the user request to the scheme in the concept knowledge graph.

[0009] Determine the scheme template based on the scheme type;

[0010] For each sub-solution in the solution template, a recommendation probability of a candidate solution is calculated according to the recommendation request, the set of meta-paths and the instance knowledge graph;

[0011] At least according to the recommendation probability of the candidate solution in each sub-solution and the solution template, a recommended solution is generated.

[0012] In some embodiments, the calculation of the recommendation probability of the candidate solution according to the recommendation request, the set of meta-paths and the instance knowledge graph comprises:

[0013] According to the recommendation request, a to-be-processed meta-path is selected from the set of meta-paths;

[0014] According to the to-be-processed meta-path, a random walk is performed in the instance knowledge graph to determine the number of hits of the candidate solution corresponding to the to-be-processed meta-path;

[0015] According to the weight of the to-be-processed meta-path and the number of hits of the candidate solution, the recommendation probability of the candidate solution is calculated.

[0016] In some embodiments, the recommendation probability of the candidate solution is calculated according to the following formula:

[0017]

[0018] Wherein, Pt is the recommendation probability of the candidate solution t, P is the set of meta-paths, p is the meta-path, Prob(q.1->t|p) is the probability of reaching the candidate solution t through the meta-path p after the random walk from q.1, n is the number of hits of the candidate solution t of the meta-path p, N is the total number of hits of each candidate solution corresponding to the meta-path p, and θp is the weight of the meta-path p.

[0019] In some embodiments, the method further comprises: if a preset weight updating condition is met, updating the weight of each meta-path; the weight updating condition comprises one of:

[0020] The update period is reached;

[0021] The number of training sample data in the instance knowledge graph increases by a preset proportion, and the training sample data is used to train the weight of the meta-path;

[0022] The concept knowledge graph is updated.

[0023] In some embodiments, the method further comprises: training the weight of each meta-path; the training of the weight of each meta-path comprises:

[0024] According to each meta-path in the meta-path set, a random walk is performed in the instance knowledge graph to determine a hit number of the candidate solution corresponding to each meta-path;

[0025] According to the current weight of each meta-path and the hit number of the candidate solution, a recommendation probability of the candidate solution corresponding to each meta-path in the meta-path set is calculated;

[0026] According to the recommendation probability of each candidate solution, the weight of each meta-path is trained.

[0027] In some embodiments, the weight of each meta-path is trained according to the following formula:

[0028]

[0029] wherein J(Θ) is a cost function, P is a meta-path set, Θ is a weight vector of a meta-path, the dimension number of the weight vector Θ is the number of meta-paths in the meta-path set P, θp is a weight value of a meta-path p, m is the number of training samples, qi is the i th user request in the training sample, ti is the candidate solution selected by the i th user request in the training sample, and Probqi->ti is a recommendation probability of selecting the candidate solution ti based on the user request qi.

[0030] In some embodiments, before the scheme template is determined according to the scheme type, the method further comprises: determining a recommendation mode according to the received scheme recommendation request;

[0031] After the scheme template is determined according to the scheme type, before the recommendation probability of the candidate solution is calculated for each sub-scheme in the scheme template according to the recommendation request, the meta-path set and the instance knowledge graph, the method further comprises: determining a processing order of each sub-scheme in the scheme template according to the recommendation mode;

[0032] The recommendation probability of the candidate solution is calculated for each sub-scheme in the scheme template according to the recommendation request, the meta-path set and the instance knowledge graph, comprising:

[0033] The recommendation probability of the candidate solution is calculated for each sub-scheme in the scheme template according to the meta-path set and the instance knowledge graph in the processing order of each sub-scheme in the scheme template.

[0034] In some embodiments, the recommendation mode comprises a first recommendation mode and a second recommendation mode, and the processing order of each sub-scheme in the scheme template is determined according to the recommendation mode, comprising:

[0035] if the recommendation mode is a second recommendation mode, determining that a processing order of a first type of sub-scheme is superior to a processing order of a second type of sub-scheme according to a dependency relationship between sub-schemes in the scheme template, wherein the first type of sub-scheme does not depend on other sub-schemes, and the second type of sub-scheme depends on other sub-schemes;

[0036] generating a recommended scheme according to the recommendation probability of the alternative scheme in each sub-scheme and the scheme template, comprises:

[0037] generating a recommended scheme according to the recommendation probability of the alternative scheme in each sub-scheme, the scheme template and the recommendation mode; wherein when the recommendation mode is a first recommendation mode, the recommended scheme is a fixed overall scheme including sub-schemes; and when the recommendation mode is a second recommendation mode, the recommended scheme includes an optimal alternative scheme of sub-schemes and other alternative schemes other than the optimal alternative scheme.

[0038] In some embodiments, after the generating a recommended scheme, the method further comprises:

[0039] receiving a user-selected scheme, and storing the user-selected scheme into the instance knowledge graph.

[0040] In some embodiments, the storing the user-selected scheme into the instance knowledge graph comprises:

[0041] preprocessing the user-selected scheme to obtain preprocessed data;

[0042] matching the preprocessed data with concepts in the concept knowledge graph to determine instances and relationships in the instance knowledge graph corresponding to the user-selected scheme;

[0043] storing the preprocessed data into the instance knowledge graph according to the instances and relationships.

[0044] In some embodiments, when the recommendation mode is a second recommendation mode, the user-selected scheme includes sub-schemes; and the method further comprises:

[0045] if the user-selected sub-scheme is not an optimal alternative scheme of the sub-scheme in the recommended scheme, and the sub-scheme is dependent on other sub-schemes, generating a recommended scheme for the dependent other sub-schemes.

[0046] In some embodiments, after the determining the meta-path set corresponding to the scheme type according to the mapping relationship between the preset scheme type and the meta-path set, the method further comprises:

[0047] obtaining associated information of the recommendation request according to the meta-path set.

[0048] The associated information is written into the instance knowledge graph.

[0049] In some embodiments, the method further includes: if the concept knowledge graph is updated, then updating the instance knowledge graph and the meta-path set.

[0050] In some embodiments, updating the metapath set includes: removing metapaths from the metapath set whose weights are less than a preset first threshold; and / or,

[0051] The metapaths in the metapath set are arranged in descending order of weight. If the number of associated information contained in the first N metapaths in the sorting is less than a preset second threshold, and the number of associated information contained in the first N+1 metapaths is greater than the second threshold, then the N+1th metapath and subsequent metapaths in the sorting are removed from the metapath set. The second threshold is the maximum value of the number of associated information obtained from multi-turn human-computer dialogue, and the associated information is information related to the recommendation request.

[0052] In some embodiments, the method further includes:

[0053] When the bearer network scheme recommendation device is activated, it updates the instance knowledge graph based on the instances and relationships in the data source, and deletes expired instances and relationships in the instance knowledge graph; and / or,

[0054] If the data source is updated, the instance knowledge graph is updated according to the updated data.

[0055] In another aspect, this disclosure also provides a bearer network scheme recommendation device, including: a receiving module, a processing module, a storage module, and a recommendation module. The storage module is used to store a concept knowledge graph for representing relationships between concepts and an instance knowledge graph for representing relationships between instances. The concepts include: user requests, scheme selection factors, and schemes. The instances are instances of each concept in the concept knowledge graph. The storage module is also used to store the mapping relationship between scheme types and meta-path sets.

[0056] The receiving module is used to receive recommendation requests;

[0057] The processing module is used to: determine the solution type based on the received recommendation request; determine the meta-path set corresponding to the solution type based on the mapping relationship between the solution type and the meta-path set, wherein the meta-path set includes multiple meta-paths, and the meta-paths represent the paths in the concept knowledge graph from the concept corresponding to the user request to the concept corresponding to the solution; and determine the solution template based on the solution type.

[0058] The recommendation module is configured to calculate a recommendation probability of a candidate solution for each sub-solution in the solution template according to the recommendation request, the set of meta-paths, and the instance knowledge graph; and generate a recommended solution according to at least the recommendation probability of the candidate solution in each sub-solution and the solution template.

[0059] In another aspect, the embodiments of the present disclosure further provide a computer device, comprising:

[0060] one or more processors;

[0061] a storage device having one or more programs stored thereon;

[0062] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for recommending a solution of a bearer network as described above.

[0063] In another aspect, the embodiments of the present disclosure further provide a computer readable medium having a computer program stored thereon, wherein the program, when executed, implements the method for recommending a solution of a bearer network as described above.

[0064] The method for recommending a solution of a bearer network provided by the embodiments of the present disclosure is applied to a device for recommending a solution of a bearer network. The device for recommending a solution of a bearer network is preconfigured with a concept knowledge graph for representing relationships between concepts and an instance knowledge graph for representing relationships between instances. The concepts include a user request, a solution selection factor, and a solution. The instances are instances of each concept in the concept knowledge graph. The method comprises the following steps: determining a solution type according to a received recommendation request; determining a set of meta-paths corresponding to the solution type according to a preconfigured mapping relationship between the solution type and the set of meta-paths. The set of meta-paths includes a plurality of meta-paths. The meta-paths represent paths between the user request and the solution in the concept knowledge graph. Determining a solution template according to the solution type. For each sub-solution in the solution template, calculating a recommendation probability of a candidate solution according to the recommendation request, the set of meta-paths, and the instance knowledge graph. Generating a recommended solution according to at least the recommendation probability of the candidate solution in each sub-solution and the solution template. The embodiments of the present disclosure convert the experience of experts into information that can be recognized by machines by using the concept knowledge graph, which can solve the problem of loss of experience in solution formulation caused by personnel flow. Moreover, the recommended solution generated based on the concept knowledge graph and the instance knowledge graph and in combination with the solution template is closer to the user's intention, thereby realizing automation and refinement of the solution formulation process. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 A schematic diagram of a solution and a sub-solution;

[0066] Figure 2A schematic diagram of a conceptual knowledge graph provided for embodiments of this disclosure;

[0067] Figure 3 A schematic diagram of an instance knowledge graph provided in an embodiment of this disclosure;

[0068] Figure 4 A schematic diagram of the conceptual knowledge graph provided in the embodiments of this disclosure in the meta-path set;

[0069] Figure 5 One of the flowcharts for the method of recommending a bearer network scheme provided in the embodiments of this disclosure;

[0070] Figure 6a This is a schematic diagram of a solution template under the first recommended mode provided in the embodiments of this disclosure;

[0071] Figure 6b This is a schematic diagram of a solution template under the second recommended mode provided in the embodiments of this disclosure;

[0072] Figure 7 A flowchart illustrating the recommended probability of alternative solutions corresponding to the meta-path provided in this embodiment of the disclosure;

[0073] Figure 8 A schematic diagram illustrating the training weights provided in this embodiment of the disclosure;

[0074] Figure 9 A second schematic flowchart illustrating the method for recommending a bearer network scheme according to an embodiment of this disclosure;

[0075] Figure 10 This is a schematic diagram illustrating the process of storing a user-selected scheme into an instance knowledge graph, as provided in an embodiment of this disclosure.

[0076] Figure 11 A schematic diagram of a conceptual knowledge graph in a specific example provided in the embodiments of this disclosure;

[0077] Figure 12 A schematic diagram of an instance knowledge graph in a specific example provided in the embodiments of this disclosure;

[0078] Figure 13 A functional schematic diagram of the bearer network scheme recommendation device provided in the embodiments of this disclosure;

[0079] Figure 14 This is a schematic diagram of the recommended device for the bearer network scheme provided in the embodiments of this disclosure. Detailed Implementation

[0080] Example embodiments will be described more fully hereinafter with reference to the accompanying drawings, in which example embodiments are shown. The example embodiments may, however, be embodied in different forms, and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided as non-limiting examples so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0081] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0082] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0083] The embodiments described herein can be described with reference to plan views and / or cross-sectional views by virtue of the fact that the disclosed embodiments can be idealized. Thus, the example illustrations can be modified according to manufacturing techniques and / or tolerances. Therefore, the embodiments are not limited to the examples illustrated in the drawings, but include modifications based on manufacturing processes. Thus, the regions illustrated in the drawings have schematic properties, and the shape of the regions shown in the drawings illustrates a specific shape of a region of an element, but is not intended to be limiting.

[0084] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted in an overly literal or overly formal sense unless expressly so defined herein.

[0085] Knowledge Graph, known as knowledge domain visualization or knowledge field mapping map in the library intelligence field, is a series of various different graphs showing the development process and structural relationship of knowledge, which uses visualization technology to describe knowledge resources and their carriers, and excavates, analyzes, constructs, draws and displays knowledge and their mutual relations.

[0086] The embodiments of the present disclosure use two kinds of knowledge graphs, which are a concept knowledge graph and an instance knowledge graph. The concept knowledge graph is a knowledge graph composed of concepts and the relationships between the concepts. The instance knowledge graph is a knowledge graph composed of instances of various concepts and the relationships between the instances. The nodes in the concept knowledge graph represent concepts, and the edges represent the relationships between the concepts. An instance of a concept in the concept knowledge graph corresponds to a node in the instance knowledge graph. As shown in Figure 2 , Figure 2 , the concept A4 in Figure 3 corresponds to the instance nodes a4.1 and a4.2 in Figure 2 . An edge in the concept knowledge graph corresponds to the category of the relationship between the corresponding instances in the instance knowledge graph. For example, Figure 3 , the edge Q->A4 in corresponds to the instances q.1->a4. and q.1->a4.2 in

[0087] .

[0088] The nodes and edges in the concept knowledge graph are manually entered, reflecting the experience of experts. The generation and update of the concept knowledge graph all need manual entry, that is, human experts need to point out the selection of the bearer network deployment scheme and what factors are related, and then summarize these factors and the relationships between the factors and manually enter them into the concept knowledge graph. The concepts in the knowledge graph include: user requests, scheme selection factors, and schemes. The scheme selection factors include but are not limited to: operators, regions, device types, customer information, network topologies, and the like. The relationships between the concepts include but are not limited to: serving, using, containing, and being located in. Figure 2 Figure 2 In the concept knowledge graph, starting from a node (such as Q in Figure 2 ), passing through one or more nodes (such as A4, A1 in Figure 2 ), and reaching another node (such as C1 in Figure 4 ), all the nodes and edges on the way are called a meta-path.

[0089] The embodiments of the present disclosure provide a bearer network scheme recommendation method, as shown in Figure 5 , the method includes the following steps:

[0090] Step 11, determining a scheme type according to the received recommendation request.

[0091] There are various types of schemes for the bearer network, which can include: deployment schemes, maintenance schemes, and cut-through schemes.

[0092] The determination manner of the scheme type can be selected by the user or extracted from the human-computer conversation through natural language processing, and the specific determination manner is not limited in the application. The person who initiates the recommendation request can be a human or a machine. For example, the local operation and maintenance personnel who hope to deploy a bearing network is a human user, and can interact with the bearing network scheme recommendation device through natural language. The bearing network scheme recommendation device collects information related to the user request through one or more interaction manners such as multiple rounds of conversation, providing a graphical interface, or providing a table for the user to fill in. If the network automatic operation and maintenance system is a machine user, for example, the machine user requests automatic expansion when congestion occurs in a VPN, and the bearing network scheme recommendation device interacts with the machine user through a rest interface.

[0093] In step 12, the meta-path set corresponding to the scheme type is determined according to the preset mapping relationship between the scheme type and the meta-path set.

[0094] The mapping relationship between the scheme type and the meta-path set is preset in the bearing network scheme recommendation device. In this step, the mapping relationship is queried according to the scheme type determined in step 11, and the meta-path set corresponding to the scheme type is obtained. As shown in Figure 4 The meta-path set includes multiple meta-paths, and the meta-path represents a path between a user request node and a scheme node in a concept knowledge graph.

[0095] In step 13, the scheme template is determined according to the scheme type.

[0096] Different types of schemes can be composed of different types of sub-schemes, so for a certain type of scheme, a scheme template is needed to describe what type of sub-scheme the scheme is composed of and how it is composed. Different scheme types have different scheme templates, Figure 6a As shown in the scheme template of the deployment scheme.

[0097] In step 14, for each sub-scheme in the scheme template, the recommendation probability of the candidate scheme is calculated according to the recommendation request, the meta-path set and the instance knowledge graph.

[0098] Each meta-path corresponds to one or more candidate schemes, for example, Figure 3 As shown in the example of the instance knowledge graph, the meta-path 2: Q-A4-A1-C1 corresponds to the candidate schemes c1.1 and c1.2. In this step, for each sub-scheme in the scheme template, the meta-path and the corresponding candidate scheme related to the recommendation request are selected, and then the PathRanking Algorithm can be used to calculate the recommendation probability of each candidate scheme, which is used as the basis for subsequent recommendation. The specific implementation manner of calculating the recommendation probability will be described in detail later.

[0099] Step 15, generating a recommended scheme according to the recommended probability of each alternative scheme in each sub-scheme and the scheme template.

[0100] In this step, when generating a recommended scheme, the alternative scheme with a large recommended probability is preferentially recommended, that is, the recommended probabilities of each alternative scheme are ranked from high to low, and the alternative scheme with a high recommended probability is preferentially recommended to the user according to the ranking. The recommended scheme of each sub-module can be determined according to the recommended probability, and then the entire recommended scheme is obtained by summarizing according to the relationship between the sub-modules defined by the scheme template.

[0101] The method for recommending a bearer network scheme provided in the embodiments of the present disclosure is applied to a bearer network scheme recommendation device, the bearer network recommendation device is preconfigured with a concept knowledge graph for representing the relationship between concepts and an instance knowledge graph for representing the relationship between instances, the concepts include a user request, a scheme selection factor and a scheme, and the instances are instances of each concept in the concept knowledge graph; the method includes: determining a scheme type according to a received recommendation request; determining a meta-path set corresponding to the scheme type according to a preconfigured mapping relationship between the scheme type and the meta-path set, the meta-path set includes a plurality of meta-paths, and the meta-paths represent paths between the user request and the scheme in the concept knowledge graph; determining a scheme template according to the scheme type; for each sub-scheme in the scheme template, calculating a recommended probability of an alternative scheme according to the recommendation request, the meta-path set and the instance knowledge graph; and generating a recommended scheme according to the recommended probability of each alternative scheme in each sub-scheme and the scheme template. The embodiments of the present disclosure convert the experience of experts into information that can be recognized by machines by using the concept knowledge graph, can solve the problem of loss of experience in scheme formulation caused by personnel flow, and can be combined with the user intent to generate a recommended scheme based on the concept knowledge graph and the instance knowledge graph and combined with the scheme template, which is more close to the user intent, thereby realizing automation and refinement of the scheme formulation process.

[0102] In some embodiments, as shown in Figure 7 The step of calculating the recommended probability of the alternative scheme according to the recommendation request, the meta-path set and the instance knowledge graph (i.e., step 14) includes the following steps:

[0103] Step 21, selecting a to-be-processed meta-path from the meta-path set according to the recommendation request.

[0104] Step 22, randomly walking in the instance knowledge graph according to the to-be-processed meta-path to determine the number of hits of the alternative scheme corresponding to the to-be-processed meta-path.

[0105] Each meta-path to be processed can correspond to one or more alternative solutions. In this step, the number of hits is determined for all alternative solutions corresponding to all meta-paths to be processed.

[0106] Step 23: Calculate the recommendation probability of the alternative solution based on the weight of the meta-path to be processed and the number of hits of the alternative solution.

[0107] In some embodiments, the recommendation probability of the alternative is calculated according to the following formula:

[0108]

[0109] Where Pt is the recommendation probability of alternative t, P is the set of metapaths, p is a metapath, and Prob(q.1->t|p) is the probability that after a random walk starting from q.1, the alternative t is reached via metapath p. n is the number of hits of the alternative scheme t for the meta-path p, N is the total number of hits of all alternative schemes corresponding to the meta-path p, and θp is the weight of the meta-path p, which can be obtained through training.

[0110] like Figure 3 As shown, in the instance knowledge graph, starting from request instance q.1, N random walks with a maximum step size of L are performed. This involves randomly selecting an edge from the adjacent edges of node q.1, then randomly moving to the opposite node of that edge, randomly selecting another edge from the adjacent edges of that opposite node, and moving to the next node, until L edges have been traversed. The nodes and edges along the path must be within the meta-path set {P}. This process is repeated N times, and the nodes and edges traversed each time are recorded, and the number of hits for each meta-path corresponding to the candidate solutions is counted.

[0111] Suppose that in N random walks, from Figure 3 Starting from Q1, passing through Figure 4 Zhongyuan Path 1 Arrival Figure 3 If the number of hits for c1.1 is n11, then the recommendation probability p1 of c1.1 is proportional to n11 multiplied by the weight of meta-path 1, i.e., p1 = n11 * 0.6. Figure 3 Starting from Q1, passing through Figure 4 Zhongyuan can be reached via paths 1, 2, and 3. Figure 3 For the keyword c1.2, the number of hits for c1.2 corresponding to metapaths 1, 2, and 3 are n21, n22, and n23 respectively. Therefore, the recommendation probability of c1.2 is p2 = n21*0.6 + n22*0.2 + n22*0.2. If p2 > p1, then c1.2 is recommended to the user first.

[0112] In some embodiments, the method further comprises the step of updating the weight of each meta-path if a preset weight updating condition is met. The weight updating condition comprises one of the following: an update period is reached; the number of training sample data in the instance knowledge graph for training the weight of the meta-path increases by a preset proportion; and the concept knowledge graph is updated.

[0113] In some embodiments, the step of training the weight of each meta-path comprises the steps of: Figure 8

[0114] Step 31, according to each meta-path in the set of meta-paths, performing random walk in the instance knowledge graph to determine the hit number of each meta-path corresponding to the candidate solution.

[0115] In this step, n times of random walk with a step length of L is performed in the instance knowledge graph according to the set of meta-paths {P} to determine the hit number of each meta-path corresponding to the candidate solution.

[0116] Step 32, according to the current weight of each meta-path and the hit number of the candidate solution, calculating the recommendation probability of each meta-path in the set of meta-paths corresponding to the candidate solution.

[0117] The current weight of each meta-path in the set of meta-paths is assumed to be wi. In this step, the recommendation probability of each meta-path corresponding to the candidate solution can be calculated according to the aforementioned recommendation probability calculation formula of the candidate solution.

[0118] Step 33, according to the recommendation probability of each candidate solution, training the weight of each meta-path.

[0119] In this step, the weight of each meta-path can be learned by using a logistic regression (softmax regression) algorithm.

[0120] In some embodiments, the weight of each meta-path is trained according to the following formula:

[0121]

[0122] wherein J(Θ) is a cost function, P is the set of meta-paths, Θ is the weight vector of the meta-path, the dimension number of the weight vector Θ is the number of meta-paths in the set of meta-paths P, θp is the weight value of the meta-path p, m is the number of training samples, qi is the i-th user request in the training sample, ti is the candidate solution selected by the i-th user request in the training sample, and Probqi->ti is the recommendation probability of selecting the candidate solution ti based on the user request qi.

[0123] ​To meet different recommendation requirements, the embodiments of the present disclosure implement two different recommendation modes. Correspondingly, in some embodiments, as shown in Figure 9 Before the step of determining the scheme template according to the scheme type (i.e., step 13), the method for recommending the bearer network scheme can further include the following steps:

[0124] Step 12', determining the recommendation mode according to the received scheme recommendation request.

[0125] After the step of determining the scheme template according to the scheme type (i.e., step 13), before the step of calculating the recommendation probability of the alternative scheme according to the recommendation request, the meta-path set and the instance knowledge graph for each sub-scheme in the scheme template (i.e., step 14), the method for recommending the bearer network scheme can further include the following steps:

[0126] Step 13', determining the processing order of each sub-scheme in the scheme template according to the recommendation mode.

[0127] Correspondingly, the step of calculating the recommendation probability of the alternative scheme according to the recommendation request, the meta-path set and the instance knowledge graph for each sub-scheme in the scheme template (i.e., step 14) includes: according to the processing order of each sub-scheme in the scheme template, calculating the recommendation probability of the alternative scheme according to the recommendation request, the meta-path set and the instance knowledge graph for each sub-scheme in the scheme template.

[0128] It should be noted that in this embodiment, the other steps are the same as those in the foregoing embodiments, and will not be described here again.

[0129] In some embodiments, the recommendation mode includes a first recommendation mode and a second recommendation mode, and the step of determining the processing order of each sub-scheme in the scheme template according to the recommendation mode (step 13') can include the following steps: if the recommendation mode is the second recommendation mode, determining that the processing order of a first type of sub-scheme is superior to the processing order of a second type of sub-scheme according to the dependency relationship between the sub-schemes in the scheme template, wherein the first type of sub-scheme does not depend on other sub-schemes, and the second type of sub-scheme depends on other sub-schemes.

[0130] In the second recommendation mode, as shown in Figure 6b The scheme template includes the dependency relationship (i.e., dependency chain) between the sub-schemes, that is, the selection of a sub-scheme can be affected by the selection of other sub-schemes in the scheme selection process, for example, the reliability scheme can be affected by the selection of the tunnel scheme. As shown in Figure 6b Sub-scheme B can only take effect when sub-scheme A takes effect, so there is a dependency relationship between sub-schemes B and A. In Figure 6bIn the shown scheme template, sub-scheme B depends on sub-scheme A, and sub-scheme E depends on sub-scheme C. When recommending the scheme, for the sub-schemes (such as sub-schemes A, D, and C) in the scheme template that do not depend on other sub-schemes, the optimal recommended scheme (optimal candidate scheme) of these sub-schemes is first determined, and then dependent sub-scheme recommendation is performed on the optimal recommended schemes of these sub-schemes, that is, sub-schemes A, D, and C need to be processed before sub-schemes B and E can be processed. In the meta path used for scheme recommendation, the dependent sub-schemes need to be in the meta path.

[0131] It should be noted that, in the first recommendation mode, as shown in Figure 6a There is no dependency (i.e., dependency chain) between the sub-schemes, the generated recommended scheme is an overall scheme including the sub-schemes, and the sub-schemes have an associated relationship defined by the scheme template. Therefore, in the process of generating the recommended scheme, the recommended schemes of the sub-modules can be processed in parallel.

[0132] In the embodiment scheme of introducing the recommendation mode, the generated recommended scheme has different forms according to different recommendation modes.

[0133] Correspondingly, the step of generating the recommended scheme according to the recommendation probability of the candidate scheme in each sub-scheme and the scheme template (i.e., step 15) can include the following steps: generating the recommended scheme according to the recommendation probability of the candidate scheme in each sub-scheme, the scheme template, and the recommendation mode. When the recommendation mode is the first recommendation mode, the recommended scheme is a fixed overall scheme including the sub-schemes; when the recommendation mode is the second recommendation mode, the recommended scheme includes the optimal candidate scheme of the sub-schemes and other candidate schemes except the optimal candidate scheme. In the recommended scheme under the second recommendation mode, the recommended schemes of the sub-schemes can be presented to the user in the form of a drop-down menu, the optimal candidate scheme is arranged at the top of the drop-down menu, and the other candidate schemes except the optimal candidate scheme can be arranged in the drop-down menu in descending order of the recommendation probability.

[0134] The first recommendation mode is a whole scheme recommendation, that is, the recommended scheme is a fixed whole scheme including sub-schemes; the second recommendation mode can recommend sub-schemes individually, and the recommended scheme includes an optimal candidate scheme of the sub-schemes and other candidate schemes except the optimal candidate scheme. In the first recommendation mode, the whole schemes are recommended to the user, and the user selects from multiple whole schemes. In this way, the recommended schemes are all existing combinations of sub-schemes. In the second recommendation mode, the user is recommended a series of combinations of sub-schemes constituting a bearer scheme, so that the user can select sub-schemes. It should be noted that this recommendation mode can generate a combination of sub-schemes that meets the user's request but has never appeared before. If the user has no knowledge of the sub-schemes constituting the bearer scheme, it is more reasonable to recommend whole schemes to the user and let the user select from multiple whole schemes. If the user has some knowledge of the sub-schemes constituting the bearer scheme, the user can be recommended sub-schemes so as to arrange and select the required sub-schemes.

[0135] For the second recommendation mode, the recommended scheme is formed by combining sub-schemes, rather than the whole scheme as a recommended result. In this way, a new combination of sub-schemes can be recommended according to the specific situation of the user, and a new scheme is formed. Thanks to the arrangement system of the bearer network, the scheme can be automatically implemented by splitting into sub-schemes, so in the recommendation of the bearer network scheme, this kind of scheme customization can be used for recommendation. From this point of view, the embodiments of the present disclosure can also be regarded as automatic generation of a whole scheme.

[0136] Further, in some embodiments, when the recommendation mode is the second recommendation mode, if the user does not select the optimal candidate scheme in the recommended scheme, and the user-selected sub-scheme is dependent on other sub-schemes, the dependent sub-schemes are re-recommended.

[0137] That is, the bearer network scheme recommendation method can further include the following steps: if the user-selected sub-scheme is not the optimal candidate scheme of the sub-scheme in the recommended scheme, and the sub-scheme is dependent on other sub-schemes, a recommended scheme is generated for the dependent other sub-schemes.

[0138] In this step, after receiving the user-selected scheme, it is first judged according to the dependency relationship in the scheme template whether the user-selected sub-scheme that is not the optimal candidate scheme is dependent on other sub-schemes. If so, the optimal candidate scheme is modified in the instance knowledge graph according to the user-selected scheme (i.e., the sub-scheme), the corresponding meta path of the affected sub-scheme is found, the recommendation probability is calculated again according to the path classification algorithm, the obtained optimal candidate scheme is assembled into the scheme according to the scheme template, and is filled into the instance knowledge graph. This step is repeatedly executed until no sub-scheme is affected, and finally the adjusted recommended scheme is presented to the user.

[0139] Further, in some embodiments, after the step of generating the recommended solution (i.e., step 14), the method can further include the following steps: receiving a user-selected solution, and storing the user-selected solution into the instance knowledge graph. In this step, the user selects one of the recommended solutions, and the knowledge synchronizer K stores the user-selected result into the instance knowledge graph.

[0140] Embodiments of the present disclosure recommend solutions to the user in descending order of recommendation probability, and the user can select the recommended solution. Since the user-selected solution is not necessarily the solution with the highest probability, the user-selected result is stored into the instance knowledge graph as a reference for subsequent recommendations and a sample for training the meta-path weight.

[0141] It should be noted that, in addition to storing the user-selected solution into the instance knowledge graph, data in the data source can also be stored into the instance knowledge graph, where the data source includes but is not limited to a bearing control system, a BSS (Business support Set), an SDN (Software Defined Network) controller, and the like.

[0142] In some embodiments, as shown in FIG. 4, the step of storing the user-selected solution into the instance knowledge graph includes the following steps: Figure 10

[0143] Step 41: Preprocessing the user-selected solution to obtain preprocessed data.

[0144] In this step, the preprocessing includes but is not limited to the following operations: using a statistical algorithm, feature extraction based on a neural network, extracting fields in a relational database according to a human-set field list, and the like.

[0145] Step 42: Matching the preprocessed data with concepts in the concept knowledge graph to determine instances and relationships in the instance knowledge graph corresponding to the user-selected solution.

[0146] In this step, the matching method includes but is not limited to the following methods: automatically setting a conversion relationship between a data type in a relational database and a concept, and specifying a concept type according to an information source.

[0147] Step 43: Storing the preprocessed data into the instance knowledge graph according to the instances and relationships.

[0148] ​In some embodiments, after determining the set of meta-paths corresponding to the scheme type according to the preset mapping relationship between the scheme type and the set of meta-paths (i.e., step 12), the method for recommending a bearer network scheme can further include the following steps: obtaining the associated information of the recommendation request according to the set of meta-paths, and writing the associated information into the instance knowledge graph.

[0149] The associated information related to the recommendation request of a user is determined by the information in the concept knowledge graph. A user request of a certain type is represented as a node Q in the concept knowledge graph as a concept, and a set of concept points {A} adjacent to the Q node (i.e., a set of points connected by lines to the Q node) and a line (or edge) between each node in the point set {A} and the Q node represent an associated information to be obtained. For example, A1 represents the concept of an operator, and the line Q-A1 represents the relationship: serving the user request Q, and the associated information to be obtained is: which operator serves the user request Q. Figure 2

[0150] The associated information to be obtained of the recommendation request is not adjacent to the user request Q in the concept knowledge graph, but each meta-path in the set of meta-paths appears in the adjacent relationship of Q, for example, Q->A4 and Q<-A1 in the above. Figure 4 After determining the scheme type, the set of meta-paths of the concept knowledge graph used in the recommendation process is determined, the associated information to be obtained is selected, the user is collected, and the collected associated information is recorded in the instance knowledge graph.

[0151] It should be noted that after the concept knowledge graph is updated, the instance knowledge graph and the meta-paths are updated, and the learning of the meta-path weight is triggered, that is, if the concept knowledge graph is updated, the instance knowledge graph and the set of meta-paths are updated.

[0152] In some embodiments, the method for recommending a bearer network scheme further includes the following steps: when the bearer network scheme recommendation device is started, updating the instance knowledge graph according to the instances and relationships in the data source, and deleting the expired instances and relationships in the instance knowledge graph; and / or, if the data source is updated, updating the instance knowledge graph according to the updated data.

[0153] ​That is, when the bearer network scheme recommendation device is just started, all knowledge synchronizers query all data of respective data sources, update new knowledge instances in the data sources to the instance knowledge graph, and delete expired nodes and edges in the instance knowledge graph. During the operation of the bearer network scheme recommendation device, the knowledge synchronizer regularly checks whether the data in the respective data source changes, and if there is a change, the changed information is synchronized to the instance knowledge graph, including adding new nodes and edges and deleting expired nodes and edges. The checking of the data source includes two ways: if the data source provides change notification, the knowledge synchronizer listens to the notification, and after receiving the change notification, the instance knowledge graph update process is started; if the data source does not provide change notification, the knowledge synchronizer will periodically obtain the data in the data source, compare it with the data in the instance knowledge graph, and update the instance knowledge graph after finding the data change.

[0154] Some meta-paths may have too low weights and thus contribute little to the scheme recommendation. If all the associated information contained in the meta-paths is asked from the user, and there is a meta-path with a very low weight, the user may be asked some questions that are not related to the user's intention, reducing the user experience. Therefore, for the associated information contained in the meta-paths with low weights, the associated information will not be included in the to-be-obtained associated information (i.e., the information that needs to be asked from the user). Correspondingly, the embodiment of the present disclosure can also update the meta-path set to remove the meta-paths with small weights.

[0155] In some embodiments, the updating the meta-path set comprises: removing a meta-path in the meta-path set whose weight is less than a preset first threshold; and / or,

[0156] The meta-paths in the meta-path set are arranged in descending order of weights, and if the number of associated information contained in the first N meta-paths in the sorting is less than a preset second threshold, and the number of associated information contained in the first N+1 meta-paths is greater than the second threshold, the first N+1 and subsequent meta-paths in the sorting are removed from the meta-path set, the second threshold is the maximum value of the number of associated information obtained from multiple rounds of human-computer dialogues, and the associated information is information related to the recommendation request.

[0157] That is, the criterion for judging whether the weight of the meta-path is low can be an absolute criterion (i.e., directly measured by a threshold value), for example, a weight threshold Pl is set, and the meta-path with a weight less than Pl is a meta-path with a weight that is too low, or a relative criterion (i.e., the weight is relatively small in the meta-path set). If the way to obtain the associated information is a multi-round human-computer dialogue, limited by the dialogue round, it is assumed that the maximum number of information collected in the multi-round dialogue (two nodes and the edge between them in the instance knowledge graph are called a triple (r, t, e), and one triple corresponds to one information) is M, the meta-paths are arranged in descending order of weight, if the number of associated information contained in the first N meta-paths is less than M, and the number of associated information contained in the first N+1 meta-paths is greater than M, then from the associated information of the N+1 meta-path, the meta-paths behind are meta-paths with low weights, and the information included in these meta-paths is no longer obtained from the user.

[0158] It should be noted that the weight judgment criterion of the meta-path described above can be implemented in a combined manner, for example: as long as one of the multiple rules is satisfied, it is considered that the weight is low.

[0159] The following will be combined Figure 13 The carrying network scheme recommendation scheme of the embodiment of the present disclosure is described. As shown in Figure 13 The user interacts with the carrying network scheme recommendation device through the interaction agent unit, that is, the carrying network scheme recommendation device receives the recommendation request (201) through the interaction agent unit.

[0160] The translator obtains the information (i.e., associated information) that needs to be collected from the user from the concept knowledge graph through 202.a, translates it into the corresponding information collection form, and the form includes but is not limited to: natural language questions, user input forms, or rest interfaces and machine languages, etc., and informs the interaction agent unit of the translation result. The above translation process performed by the translator includes but is not limited to: configuring the relationship between the mapping file and the corresponding problem or interface through the corresponding method, automatically converting the concept knowledge graph into text by using the NPL method, etc.

[0161] The interaction agent unit informs the translator of the information collected from the user through 202.b, and the translator stores the relevant information into the instance knowledge graph through 203. The data storage process in the instance knowledge graph is a process of establishing a series of nodes and edges. As shown in Figure 3As shown, for the information that "the operator serving the user request q.1 is a1.1", if a1.1 already exists in the instance knowledge graph, only a line representing the service relationship needs to be established between a1.1 and q.1 (i.e., a relationship is established). If a1.1 does not exist in the instance knowledge graph, a node representing a1.1 is first established, and then a line between a1.1 and q.1 is established. After the instance knowledge graph stores the information collected by the interactive agent unit from the user, the translator informs the recommender of the type of the recommended scheme through 204.a, and starts the calculation process of the scheme recommendation.

[0162] The recommender obtains the meta-paths of the concept knowledge graph through 204.b, and establishes and stores the mapping relationship between the scheme type and the meta-path set. After the recommender generates the recommended scheme, the recommended result is presented to the interactive agent unit through 205. After the user determines the final scheme, the scheme is handed over to the automatic management and control system through the interactive agent unit for execution. The automatic management and control system can include a series of components such as orchestrator and controller for automatically managing and controlling the network, and can decompose the scheme layer by layer and automatically deploy.

[0163] The knowledge synchronizer is responsible for storing various information into the instance knowledge graph. There can be multiple knowledge synchronizers, one knowledge synchronizer corresponding to one information source. Different information sources need to be matched with the knowledge synchronizer K to realize the synchronization of knowledge. For example, the user selects one of the recommended results through 201, and the interactive agent unit delivers the user-selected result to the knowledge synchronizer K through 105, and the knowledge synchronizer K stores the user-selected result into the instance knowledge graph. The database in the automatic management and control system records the actual topology information in the network, and the knowledge synchronizer K+1 converts the topology configuration, device type and other information and stores them into the instance knowledge graph.

[0164] The weight of the meta-path is obtained through machine learning training, which is realized by the weight updater, and the training result is pushed to the recommender through 104. The weight of the meta-path can be updated (103) using the existing data in the instance knowledge graph, and the update strategy is as follows: regular update, or concept knowledge graph update, or the number of training sample data in the instance knowledge graph increases by a preset proportion.

[0165] In order to clearly illustrate the scheme of the embodiments of the present disclosure, the following will be described in detail in combination with Figure 11 , 12 , 13 through a specific embodiment.

[0166] In the initialization phase, before the bearer network scheme recommendation device starts, the human experts write the concepts related to the recommended scheme and the relationships between them into a text, and the bearer network scheme recommendation device automatically enters these information into the concept knowledge graph when it starts, obtaining, for example, Figure 10The concept knowledge graph shown.

[0167] After the system is started, the knowledge synchronizer performs knowledge synchronization, and information is entered into the instance knowledge graph, obtaining an instance knowledge graph as shown in Figure 11

[0168] If the meta-path weight in the recommender is not updated or has never been trained (first start), the weight updater uses the existing user request instances in the instance knowledge graph to train the weight of the meta-path, and pushes it to the recommender for weight updating. For example, the meta-path weight of step L <= 2 is trained as follows:

[0169] Meta-path a: user request -> business user -> VPN solution, weight 0.4

[0170] Meta-path b: user request -> business scenario -> VPN solution, weight 0.6

[0171] The above meta-path a represents recommending the network solution for the current request according to the network solution used by the business user to which the user request belongs, with a weight of 0.4; and meta-path b represents recommending the network solution for the current request according to the network solution used by the business scenario to which the user request belongs, with a weight of 0.6.

[0172] In this specific example, the default recommendation mode is the second recommendation mode.

[0173] The human user proposes to the interactive agent unit to deploy an enterprise private network for a certain large customer U2 by using human-computer dialogue. The interactive agent unit analyzes that the solution type is a network deployment solution, and then finds out the corresponding meta-path set {a, b}, determines the associated information that needs to be collected from the human user, for example, "what is the name of the large customer", extracts the associated information "business user U2" from the human user's interaction, and delivers the associated information to the translator. The translator queries whether there is information about the company in the instance knowledge graph, if there is, the next question is asked; if there is not, the translator further asks for information according to the information associated with the "customer" concept in the meta-path, and then completes the collection of associated information from the human user through multiple rounds of dialogue.

[0174] The translator enters the information collected by the interactive agent unit associated with the user request into the instance knowledge graph, such as "request 2 of U2" as shown in Figure 12 and notifies the recommender to recommend a solution. The recommender determines that it is currently in the second recommendation mode, so it finds the solution template of the enterprise private network (as shown in Figure 5 ), and all sub-solutions in the template have no dependency, so it starts recommending each sub-solution.

[0175] ​The recommendation process of the VPN sub-solution is as follows:

[0176] The recommender performs 10 random walks in the instance knowledge graph according to the corresponding meta path {a, b}, and then counts the number of hits of the candidate solutions in each meta path as follows:

[0177] Meta path a: user request -> business user -> network solution (weight 0.4), L2+L3 hit 5 times, VPWS hit 5 times;

[0178] Meta path b: user request -> business scenario -> network solution (weight 0.6), VPLS hit 3 times, L3VPN hit 3 times, VPWS hit 4 times;

[0179] Combined with the weights of the meta paths, the recommendation probabilities of each candidate solution are calculated as follows:

[0180] The recommendation probability of L2+L3: 5 / 10*0.4+0*0.6=0.2;

[0181] The recommendation probability of VPLS: 0*0.4+3 / 10*0.6=0.18;

[0182] The recommendation probability of VPWS: 5 / 10*0.4+4 / 10*0.6=0.44;

[0183] The recommendation probability of L3VPN: 0*0.4+3 / 10*0.6=0.18;

[0184] The recommender prioritizes the candidate solutions with high probabilities in descending order of probability, that is, the recommendation probability of VPWS is 44% and ranks first, the recommendation probability of L2+L3 is 20% and ranks second, and the recommendation probabilities of L3VPN and VPLS are both 18% and rank last.

[0185] The recommendation process of other sub-solutions (such as the tunnel sub-solution) is similar to that of the VPN sub-solution. After completing the recommendation of each sub-solution, the optimal recommended sub-solution set is found, and the final recommended solution is constructed according to the solution template.

[0186] The interactive agent unit presents the recommended solution to the user. If the user makes an adjustment selection for one of the sub-solutions, for example, the user selects VPLS to replace the VPWS recommended by the system, the interactive agent unit pushes the user's selection result to the knowledge synchronizer, which records the user's selection result in the instance knowledge graph, and at the same time, the adjusted solution is handed over to the automated management and control system for deployment.

[0187] The following takes "recommending network expansion and contraction solution to the automated closed-loop system" as an example to explain the solution of the embodiment of the present disclosure in detail.

[0188] As a machine user, the automatic management and control system detects congestion of a VPLS service opened for "XX Company", and cannot meet the requirements of SLA (service level agreement), and calls a rest interface to request expansion to an interactive agent unit. The interactive agent judges that the recommended scheme type is an expansion scheme, and then finds the corresponding meta path, determines the associated information to be collected from the automatic management and control system, and the translator converts the corresponding information into a rest interface. The interactive agent unit calls the rest interface for query, and enters the collected associated information into the instance knowledge graph through the translator, and then informs the recommender to recommend a scheme.

[0189] The recommender performs random walk in the instance knowledge graph according to the corresponding meta path of the expansion scheme, counts the hit number of the corresponding alternative scheme of each meta path, calculates the recommendation probability of each expansion scheme by using the weight of each meta path, finds the scheme with the highest probability as the recommended scheme, and delivers it to the automatic management and control system for execution.

[0190] Based on the same technical concept, the embodiment of the disclosure also provides a bearer network scheme recommendation device, as shown in Figure 14 The bearer network scheme recommendation device includes a receiving module 101, a processing module 102, a storage module 103, and a recommendation module 104. The storage module 103 is used to store a concept knowledge graph for representing the relationship between concepts and an instance knowledge graph for representing the relationship between instances. The concepts include user requests, scheme selection factors, and schemes. The instances are instances of each concept in the concept knowledge graph. The storage module 103 is also used to store a mapping relationship between a scheme type and a meta path set.

[0191] The receiving module 101 is used to receive a recommendation request.

[0192] The processing module 102 is used to determine a scheme type according to the received recommendation request, determine a meta path set corresponding to the scheme type according to the mapping relationship between the scheme type and the meta path set, and determine a scheme template according to the scheme type. The meta path set includes multiple meta paths, and the meta path represents a path between a concept corresponding to the user request and a concept corresponding to the scheme in the concept knowledge graph.

[0193] The recommendation module 104 is used to calculate the recommendation probability of an alternative scheme according to the meta path set and the instance knowledge graph for each sub-scheme in the scheme template, and generate a recommended scheme according to the recommendation probability of the alternative scheme in each sub-scheme and the scheme template.

[0194] In some embodiments, the recommendation module 104 is configured to select a to-be-processed meta-path from the set of meta-paths according to the recommendation request, perform random walk in the instance knowledge graph according to the to-be-processed meta-path to determine a hit number of the candidate solution corresponding to the to-be-processed meta-path, and calculate a recommendation probability of the candidate solution according to the weight of the to-be-processed meta-path and the hit number of the candidate solution.

[0195] In some embodiments, the recommendation module 104 is configured to calculate the recommendation probability of the candidate solution according to the following formula:

[0196]

[0197] wherein Pt is the recommendation probability of the candidate solution t, P is the set of meta-paths, p is a meta-path, Prob(q.1->t|p) is a probability of reaching the candidate solution t through the meta-path p after random walk from q.1, n is the hit number of the candidate solution t of the meta-path p, N is a total number of hits of each candidate solution corresponding to the meta-path p, and θp is the weight of the meta-path p.

[0198] In some embodiments, if a preset weight updating condition is met, the weights of the meta-paths are updated, and the weight updating condition includes one of the following:

[0199] an update cycle is reached,

[0200] a number of training sample data in the instance knowledge graph increases by a preset proportion, and the training sample data is used to train the weights of the meta-paths,

[0201] the concept knowledge graph is updated.

[0202] In some embodiments, the bearer network solution recommendation apparatus further includes a training module configured to perform random walk in the instance knowledge graph according to each meta-path in the set of meta-paths to determine a hit number of a candidate solution corresponding to each meta-path, calculate a recommendation probability of the candidate solution corresponding to each meta-path in the set of meta-paths according to a current weight of each meta-path and the hit number of the candidate solution, and train the weights of the meta-paths according to the recommendation probabilities of the candidate solutions.

[0203] The training module is configured to train the weights of the meta-paths according to the following formula:

[0204]

[0205] wherein J(Θ) is a cost function, P is a set of meta-paths, Θ is a weight vector of the meta-paths, a dimension number of the weight vector Θ is a number of meta-paths in the set of meta-paths P, θ p is a weight value of a meta-path p, m is a number of training samples, q i is an i th user request in the training samples, t i is a candidate selected by the i th user request in the training samples, and Prob q i -> t i is a recommendation probability of selecting the candidate t i based on the user request q i.

[0206] In some embodiments, the processing module 102 is further configured to, before determining the scheme template according to the scheme type, determine a recommendation mode according to the received scheme recommendation request; and after determining the scheme template according to the scheme type, before calculating the recommendation probability of the candidate for each sub-scheme in the scheme template according to the recommendation request, the set of meta-paths and the instance knowledge graph, determine a processing order of the sub-schemes in the scheme template according to the recommendation mode.

[0207] The recommendation module 104 is configured to, according to the processing order of the sub-schemes in the scheme template, calculate the recommendation probability of the candidate for each sub-scheme in the scheme template according to the recommendation request, the set of meta-paths and the instance knowledge graph.

[0208] In some embodiments, the recommendation mode includes a first recommendation mode and a second recommendation mode, and the processing module 102 is configured to, if the recommendation mode is the second recommendation mode, determine that the processing order of a first type of sub-scheme is superior to the processing order of a second type of sub-scheme according to a dependency relationship between the sub-schemes in the scheme template, wherein the first type of sub-scheme does not depend on other sub-schemes, and the second type of sub-scheme depends on other sub-schemes.

[0209] The recommendation module 104 is configured to generate a recommended scheme according to the recommendation probability of the candidate in each sub-scheme, the scheme template and the recommendation mode; wherein when the recommendation mode is the first recommendation mode, the recommended scheme is a fixed overall scheme including the sub-schemes; and when the recommendation mode is the second recommendation mode, the recommended scheme includes an optimal candidate of the sub-schemes and other candidates other than the optimal candidate.

[0210] In some embodiments, the receiving module 101 is further configured to receive a user-selected scheme.

[0211] The processing module 102 is further configured to store the user-selected scheme into the instance knowledge graph.

[0212] In some embodiments, the processing module 102 is configured to pre-process the user-selected scheme to obtain pre-processed data, match the pre-processed data with concepts in the concept knowledge graph to determine instances and relationships in the instance knowledge graph corresponding to the user-selected scheme, and store the pre-processed data into the instance knowledge graph according to the instances and relationships.

[0213] In some embodiments, when the recommendation mode is a second recommendation mode, the user-selected scheme includes a sub-scheme; and the recommendation module 104 is further configured to, if the user-selected sub-scheme is not the optimal candidate sub-scheme in the recommended scheme, and the sub-scheme is dependent on other sub-schemes, generate a recommended scheme for the dependent other sub-schemes.

[0214] In some embodiments, the processing module 102 is further configured to, after determining the meta-path set corresponding to the scheme type according to the mapping relationship between the preset scheme type and the meta-path set, acquire associated information of the recommendation request according to the meta-path set, and write the associated information into the instance knowledge graph.

[0215] In some embodiments, the bearer network scheme recommendation apparatus further includes an updating module configured to, when the concept knowledge graph is updated, update the instance knowledge graph and the meta-path set.

[0216] In some embodiments, the updating module is configured to remove a meta-path in the meta-path set whose weight is less than a preset first threshold value, and / or arrange each meta-path in the meta-path set in descending order of weight, and if the number of associated information contained in the first N meta-paths in the order is less than a preset second threshold value, and the number of associated information contained in the first N+1 meta-paths is greater than the second threshold value, remove the (N+1)th and subsequent meta-paths in the order from the meta-path set, the second threshold value being the maximum value of the number of associated information acquired from multiple rounds of human-computer dialogues, and the associated information being information related to the recommendation request.

[0217] In some embodiments, the updating module is further configured to, when the bearer network scheme recommendation apparatus is started, update the instance knowledge graph according to instances and relationships in a data source, and delete expired instances and relationships in the instance knowledge graph; and / or, when data update occurs in the data source, update the instance knowledge graph according to the updated data.

[0218] The embodiments of the present disclosure further provide a computer device, comprising: one or more processors and a storage device; wherein the storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method for recommending a bearer network solution as provided in the foregoing embodiments.

[0219] The embodiments of the present disclosure further provide a computer readable medium, which stores a computer program, and when the computer program is executed, the method for recommending a bearer network solution as provided in the foregoing embodiments is implemented.

[0220] Those skilled in the art can understand that all or some of the steps in the above disclosed method and the functions of the functional modules / units in the device can be implemented by software, firmware, hardware, or a combination thereof. In the hardware implementation, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component can have multiple functions, or one function or step can be performed by several physical components in cooperation. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on computer readable media, which can include computer storage media (or non-transitory media) and communication media (or transitory media). As is well known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media typically includes computer readable instructions, data structures, program modules or other data in modulated data signals such as carrier waves or other transport mechanisms, and can include any information delivery medium.

[0221] Example embodiments have been disclosed herein and, although specific terms are employed, they are used in a generic and descriptive sense only and not for purposes of limitation. In some instances, it will be apparent to those skilled in the art that features, characteristics or / and elements described in connection with a particular embodiment can be used in conjunction with other embodiments unless otherwise explicitly stated. Accordingly, it will be understood that various changes in form and details can be made without departing from the scope of the present application as set forth in the appended claims.

Claims

1. A method for recommending a bearer network scheme, characterized in that, The method is applied to a bearer network scheme recommendation device. The bearer network recommendation device has a pre-set concept knowledge graph representing relationships between concepts and an instance knowledge graph representing relationships between instances. The concepts include: user requests, scheme selection factors, and schemes. The instances are instances of each concept in the concept knowledge graph. The method includes: Determine the solution type based on the received recommendation request; The meta-path set corresponding to the scheme type is determined according to the preset mapping relationship between scheme type and meta-path set. The meta-path set includes multiple meta-paths, and the meta-paths represent the paths from the user request to the scheme in the concept knowledge graph. Determine the scheme template based on the scheme type; For each sub-scheme in the scheme template, the recommendation probability of the alternative scheme is calculated based on the recommendation request, the meta-path set, and the instance knowledge graph. A recommended solution is generated based at least on the recommendation probability of the alternative solutions described in each sub-solution and the solution template; The step of calculating the recommendation probability of alternative solutions based on the recommendation request, the meta-path set, and the instance knowledge graph includes: Based on the recommendation request, select the meta-path to be processed from the meta-path set; Based on the meta path to be processed, a random walk is performed in the instance knowledge graph to determine the number of hits of the alternative solutions corresponding to the meta path to be processed. The recommendation probability of the alternative solution is calculated based on the weight of the meta-path to be processed and the number of hits of the alternative solution.

2. The method as described in claim 1, characterized in that, The recommendation probability of the alternative is calculated using the following formula: Where Pt is the recommendation probability of alternative t, P is the set of metapaths, p is a metapath, and Prob(q.1->t|p) is the probability that after a random walk starting from q.1, the alternative t is reached via metapath p. n is the number of hits of the alternative scheme t for the meta-path p, N is the total number of hits of all alternative schemes corresponding to the meta-path p, and θp is the weight of the meta-path p.

3. The method as described in claim 1, characterized in that, Also includes: If the preset weight update conditions are met, then the weights of each meta-path are updated; the weight update conditions include one of the following: The update cycle has arrived; The number of training sample data in the instance knowledge graph is increased by a preset ratio, and the training sample data is used to train the weights of the meta-path. The concept knowledge graph has been updated.

4. The method as described in claim 1, characterized in that, Also includes: Train the weights of each metapath; The weights of each metapath during training include: Based on each meta-path in the meta-path set, a random walk is performed in the instance knowledge graph to determine the number of hits for each meta-path corresponding to the alternative solutions. Based on the current weight of each metapath and the number of hits of the alternative solutions, the recommendation probability of the alternative solutions corresponding to each metapath in the metapath set is calculated. The weights of each metapath are trained based on the recommendation probabilities of each alternative.

5. The method as described in claim 4, characterized in that, The weights of each metapath are trained using the following formula: Where J(Θ) is the cost function, P is the set of metapaths, Θ is the weight vector of the metapath, the dimension of the weight vector Θ is the number of metapaths in the set of metapaths P, θp is the weight value of metapath p, m is the number of training samples, qi is the i-th user request in the training samples, ti is the alternative selected by the i-th user request in the training samples, and Prob(qi->ti|p) is the recommendation probability of selecting alternative ti based on user request qi.

6. The method as described in claim 1, characterized in that, Before determining the scheme template based on the scheme type, the method further includes: determining a recommendation mode based on the received scheme recommendation request; After determining the scheme template according to the scheme type, and before calculating the recommendation probability of the alternative scheme for each sub-scheme in the scheme template based on the recommendation request, the meta-path set, and the instance knowledge graph, the method further includes: determining the processing order of each sub-scheme in the scheme template according to the recommendation mode; For each sub-scheme in the scheme template, the recommendation probability of the candidate scheme is calculated based on the recommendation request, the meta-path set, and the instance knowledge graph, including: According to the processing order of each sub-scheme in the scheme template, for each sub-scheme in the scheme template, the recommendation probability of the alternative scheme is calculated based on the recommendation request, the meta-path set, and the instance knowledge graph.

7. The method as described in claim 6, characterized in that, The recommendation mode includes a first recommendation mode and a second recommendation mode. Determining the processing order of each sub-scheme in the scheme template according to the recommendation mode includes: If the recommendation mode is the second recommendation mode, then based on the dependency relationship between the sub-solutions in the solution template, the processing order of the first type of sub-solution is determined to be better than the processing order of the second type of sub-solution, wherein the first type of sub-solution does not depend on other sub-solutions, and the second type of sub-solution depends on other sub-solutions; The step of generating a recommended solution based at least on the recommendation probability of the alternative solutions in each sub-solution and the solution template includes: A recommended solution is generated based on the recommendation probability of the alternative solutions in each sub-solution, the solution template, and the recommendation mode; wherein, when the recommendation mode is the first recommendation mode, the recommended solution is a fixed overall solution including the sub-solutions; when the recommendation mode is the second recommendation mode, the recommended solution includes the optimal alternative solution of the sub-solutions and other alternative solutions besides the optimal alternative solution.

8. The method as described in claim 7, characterized in that, After generating the recommendation scheme, the method further includes: Receive the solution selected by the user and store the solution selected by the user in the instance knowledge graph.

9. The method as described in claim 8, characterized in that, The step of storing the user-selected solution in the instance knowledge graph includes: The user-selected scheme is preprocessed to obtain preprocessed data; The preprocessed data is matched with the concepts in the concept knowledge graph to determine the instances and relationships in the instance knowledge graph corresponding to the scheme selected by the user. Based on the instances and relationships, the preprocessed data is stored in the instance knowledge graph.

10. The method as described in claim 8, characterized in that, When the recommendation mode is the second recommendation mode, the user-selected scheme includes sub-schemes; the method further includes: If the sub-solution selected by the user is not the optimal alternative among the sub-solutions in the recommended solution, and the sub-solution is depended upon by other sub-solutions, then a recommended solution is generated for the other dependent sub-solutions.

11. The method according to any one of claims 1-10, characterized in that, After determining the metapath set corresponding to the scheme type based on the preset mapping relationship between scheme types and metapath sets, the method further includes: The association information of the recommendation request is obtained based on the meta-path set; The associated information is written into the instance knowledge graph.

12. The method as described in claim 1, characterized in that, The method further includes: if the concept knowledge graph is updated, then the instance knowledge graph and the meta-path set are updated.

13. The method as described in claim 12, characterized in that, Updating the metapath set includes: removing metapaths from the metapath set whose weights are less than a preset first threshold; and / or, The metapaths in the metapath set are arranged in descending order of weight to obtain a sequence. If the number of associated information contained in the first N metapaths in the sequence is less than a preset second threshold, and the number of associated information contained in the first N+1 metapaths is greater than the second threshold, then the N+1th metapath and subsequent metapaths in the sequence are removed from the metapath set. The second threshold is the maximum value of the number of associated information obtained from multi-turn human-computer dialogue, and the associated information is information related to the recommendation request.

14. The method as described in claim 1, characterized in that, Also includes: When the bearer network scheme recommendation device is activated, it updates the instance knowledge graph based on the instances and relationships in the data source, and deletes expired instances and relationships in the instance knowledge graph; and / or, If the data source is updated, the instance knowledge graph is updated according to the updated data.

15. A device for recommending a bearer network scheme, comprising: The system comprises a receiving module, a processing module, a storage module, and a recommendation module. The storage module is used to store a concept knowledge graph representing relationships between concepts and an instance knowledge graph representing relationships between instances. The concepts include user requests, solution selection factors, and solutions. The instances are instances of each concept in the concept knowledge graph. The system also stores the mapping relationship between solution types and meta-path sets. The receiving module is used to receive recommendation requests; The processing module is used to: determine the solution type based on the received recommendation request; determine the meta-path set corresponding to the solution type based on the mapping relationship between the solution type and the meta-path set, wherein the meta-path set includes multiple meta-paths, and the meta-paths represent the paths in the concept knowledge graph from the concept corresponding to the user request to the concept corresponding to the solution; and determine the solution template based on the solution type. The recommendation module is used to calculate the recommendation probability of candidate solutions for each sub-solution in the solution template, based on the recommendation request, the meta-path set, and the instance knowledge graph; and to generate a recommended solution based at least on the recommendation probabilities of the candidate solutions in each sub-solution and the solution template. The step of calculating the recommendation probability of candidate solutions based on the recommendation request, the meta-path set, and the instance knowledge graph includes: selecting a meta-path to be processed from the meta-path set based on the recommendation request; performing a random walk in the instance knowledge graph based on the meta-path to be processed to determine the number of hits of candidate solutions corresponding to the meta-path to be processed; and calculating the recommendation probability of the candidate solution based on the weight of the meta-path to be processed and the number of hits of the candidate solutions.

16. A computer device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the bearer network scheme recommendation method as described in any one of claims 1-14.

17. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed, it implements the bearer network scheme recommendation method as described in any one of claims 1-14.

Images