CN117972231A - RPA project recommendation method, storage medium and electronic equipment - Google Patents
RPA project recommendation method, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN117972231A CN117972231A CN202410383490.1A CN202410383490A CN117972231A CN 117972231 A CN117972231 A CN 117972231A CN 202410383490 A CN202410383490 A CN 202410383490A CN 117972231 A CN117972231 A CN 117972231A
- Authority
- CN
- China
- Prior art keywords
- user
- project
- representation
- rpa
- item
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000003860 storage Methods 0.000 title claims abstract description 9
- 230000006870 function Effects 0.000 claims abstract description 58
- 238000004519 manufacturing process Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 6
- 238000005457 optimization Methods 0.000 claims abstract description 5
- 238000005070 sampling Methods 0.000 claims description 23
- 230000003993 interaction Effects 0.000 claims description 21
- 238000004220 aggregation Methods 0.000 claims description 15
- 230000002776 aggregation Effects 0.000 claims description 15
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 4
- 239000013598 vector Substances 0.000 claims description 4
- 238000013527 convolutional neural network Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012856 packing Methods 0.000 claims description 3
- 230000004931 aggregating effect Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 230000006399 behavior Effects 0.000 description 6
- 238000004801 process automation Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 230000003252 repetitive effect Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000003278 mimic effect Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an RPA project recommending method, a storage medium and electronic equipment, belonging to the technical field of data processing, wherein the method comprises the following steps: constructing a user project bipartite graph learning user representation; constructing a user network by using the user project bipartite graph; extracting RPA item representations in an RPA knowledge graph using a knowledge graph convolution network framework based on a user's mixed attention coefficients for entity relationships and entity preferences to construct a bipartite graph loss functionSimilarity constraint loss function between user representationsLearning a loss function of an RPA item representation from a knowledge graph; Bond loss function、、And constructing a final loss function training optimization recommendation model, and deploying the obtained optimal recommendation model in a production environment to recommend RPA (remote procedure A) projects for users. The invention captures the high-order collaborative signal in the two charts of the user project to strengthen the user representation, improves the accuracy of recommendation, effectively enhances the propagation of user preference in the knowledge graph, and is beneficial to learning personalized project representation.
Description
Technical Field
The invention belongs to the technical field of data processing, and particularly relates to an RPA project recommending method, a storage medium and electronic equipment.
Background
The recommendation system is an intelligent system based on user behavior data, analyzes interest preferences of users according to the historical behavior data of the users, recommends relevant contents to the users through personalized preferences, and improves user experience and satisfaction. Conventional recommendation methods typically suffer from sparse interaction data and cold starts, and thus researchers have attempted to utilize some additional information to improve the performance of the recommendation system. For example, researchers have discovered that information such as the nature of users and items, the network of users, etc. can be used to assist in recommendations. Such ancillary information may help the recommendation system better understand user interests and preferences, particularly for new users or new items, to more accurately recommend items.
The knowledge graph is a directional heterogeneous information network containing a large number of entities and relations, and is auxiliary information commonly used in recommendation systems. Essentially, a knowledge graph is a graphical database for storing and representing knowledge, whose nodes represent entities or attributes, and edges represent relationships between the entities or attributes. Potential relations between projects can be found by mining semantic relations between entities in the knowledge graph, so that more accurate recommendation results are provided for users. And secondly, the knowledge graph contains rich relation information, and the interest range of the user can be reasonably expanded based on various relations of the knowledge graph, so that the diversity of recommended items is increased. Finally, by linking the items of the user history interaction with the items to be recommended on the knowledge graph, a recommendation path is formed, so that the user can more convince the recommendation result, and the interpretability is provided for the recommendation result. In general, recommendation methods based on knowledge maps can be classified into the following three categories: (1) embedding-based methods. The method generally uses a knowledge graph embedding model to express entities and relations and is used for enriching the original project and semantic information of users; (2) a path-based method. The method mainly considers path information in a knowledge graph to assist recommendation, specifically, firstly, extracting similarity between target nodes by using a plurality of predefined element paths, and finding a plurality of paths from the knowledge graph to establish connection between two entities; recommending the entity with higher similarity with the target node to the user; (3) a method based on a graph neural network. A graph neural network is a neural network model that automatically extracts features from a graph structure, which aggregates and propagates information about nodes and edges in the graph, and extracts representative vectors of the nodes therefrom. The unified graph neural network is used for modeling the user interaction information and the knowledge graph, so that the collaborative signal of the user interaction with the project can be encoded clearly, the representation of the user and the project can be enhanced through the propagation process of the neural network, and the recommendation precision can be improved.
Robotic process automation (Robotic Process Automation, RPA) is a technique by which repetitive tasks performed by a human user on a computer are simulated and performed by a computer program. RPA technology typically uses a software robot or "robot worker" to automatically perform these tasks, which may mimic the behavior of a human user, such as clicking on a button in a Graphical User Interface (GUI), filling in a form, processing data, and so forth. Since RPA can automatically perform cumbersome, repetitive work tasks, it has been widely used in the financial, insurance, healthcare, retail, and manufacturing industries to improve productivity, reduce errors, and save time and cost.
Robotic Process Automation (RPA) generates a large amount of data during the execution of tasks, including but not limited to: operation text information, operation interface image information, behavior intention information among operations and operation log information. These data can be used to monitor and evaluate the performance of the robot and provide valuable information for optimization of the business process. Meanwhile, the data can be analyzed through artificial intelligence, machine learning and other technologies, so that more efficient and accurate robot process automation can be realized. Patent application: the method, the device, the server and the storage medium for establishing the RPA knowledge graph are used for automatically executing the acquisition operation based on the RPA to acquire text information of the software to be tested used by a user and generating the knowledge graph based on the association relation between the operation information and the text information. Patent application: the knowledge graph construction method combining the RPA and the AI is provided, a triplet information set required in knowledge graph construction is obtained through an RPA system and a triplet extraction technology, and the obtained triplet information is combined to construct a corresponding knowledge graph, so that the accuracy of knowledge graph construction is improved. Patent application CN112232915A: a commodity recommendation method combining RPA and AI is provided, which is used for recalling candidate commodities for recommendation according to different recall strategies adapted by different shopping scene information. Most of the existing works are to construct a knowledge graph based on information generated in an RPA implementation flow, or directly adapt different recommendation scenes to help recommendation for users by utilizing the RPA.
In recent years, a recommendation algorithm oriented to a knowledge graph has been remarkably successful, but some problems still exist. First, most existing knowledge-based recommendations focus on learning item representations in the knowledge-graph, and do not fully exploit the high-order collaboration signals between user item interactions. For example, KGCN and KGCN-LS learn item representations based on knowledge-graph, fail to model user preferences with information into user interaction data. In addition, some models (e.g., KGAT and CKAN) construct the user-item bipartite graph and the knowledge graph as a unified collaborative graph, but still perform high-order aggregate propagation only on the knowledge graph, and do not fully exploit the high-order collaborative signals between user-item interactions. Secondly, most of the existing researches use relation awareness attention to distribute weights in neighborhood aggregation, and the method cannot fully spread user preference in a knowledge graph with less relation types, so that personalized item representations are not easy to learn.
Disclosure of Invention
The invention aims at: in order to solve the problems that the prior recommendation algorithm does not fully utilize high-order collaborative signals between user item interactions, can not fully propagate user preferences and is unfavorable for learning personalized item representations, the invention provides an RPA item recommendation method, which comprises the following steps:
s1, acquiring user and project information, constructing a user project bipartite graph, and learning user representation by using the user project bipartite graph;
s2, constructing a user network by utilizing a user project bipartite graph based on the user representation;
S3, acquiring an RPA knowledge graph, and extracting an RPA project representation from the RPA knowledge graph by using a knowledge graph convolution network framework based on a mixed attention coefficient of a user on the relationship and entity preference in the RPA knowledge graph;
S4, constructing a recommendation model, wherein the recommendation model comprises a knowledge graph convolution network framework of the user network and a mixed attention coefficient based on the entity relation and entity preference of the user; constructing a user project bipartite graph loss function by setting model convergence mode ; Based on the user network, constructing a similarity constraint loss function/>, between user representations; Obtaining a loss function/>, based on the user representation and the RPA item representation, of learning the RPA item representation in the knowledge graph; Binding loss function/>、/>、Constructing a final loss function training recommendation model to obtain a final recommendation model;
And S5, deploying the final recommendation model in a production environment, inputting a user project bipartite graph and an RPA knowledge graph of the production environment, obtaining user representation and RPA project representation of the production environment, and obtaining a prediction result by solving an inner product of the user representation and the RPA project representation of the production environment, wherein the prediction result is used for recommending the RPA project.
Further, in step S2, based on the user representation, constructing the user network using the user project bipartite graph is specifically:
constructing a user network G= (A, X) by using a user project bipartite graph, wherein A is an adjacency matrix of user nodes in the user project bipartite graph, and represents whether edge connection exists between users, N user representations are learned for two graphs of user items.
Further, constructing a user project bipartite graph loss function by setting a model convergence modeThe method specifically comprises the following steps:
Setting a model convergence state:
Wherein, Representation of representation user u,/>Representing the representation of item i, N (u) representing the historical interaction items of user u,Constraint coefficient representing user u and item i,/>Representing user node degree,/>, in user project bipartite graphRepresenting the node degree of the item in the two-part diagram of the user item;
constraint loss on the user project bipartite graph is as follows:
Wherein, Representing constraint loss on the user project bipartite graph; /(I)Representing a positive sampling set, wherein edges exist between user nodes and project nodes in the positive sampling set; /(I)Representing a negative sampling set, wherein no edges exist between user nodes and project nodes in the negative sampling set; /(I)Representing constraint coefficients of user u and item j, σ represents sigmod functions,/>A representation of a representation item j;
using BCE loss as the primary optimization objective for learning user representations in a user project bipartite graph, the BCE loss for a user project bipartite graph is as follows:
Wherein, BCE loss representing user project bipartite graph;
The user project bipartite graph loss function is as follows:
Wherein, A graph loss function for the user project bipartite.
Further, based on the user network, a similarity constraint loss function between user representations is constructedThe method specifically comprises the following steps:
the social similarity among users is calculated by adopting cosine similarity as follows:
Wherein, For/>And/>Cosine similarity between/(Representing the ith user representation,/>Representing the j-th user representation,/>Representing the modular length of the vector;
the similarity constraint loss between the constructed user representations is as follows:
Wherein, Constraint loss function for similarity between user representations,/>Representing neighbor nodes of the ith user in the user network, and n represents the total number of users.
Further, in step S3, the RPA item representation is extracted from the RPA knowledge graph using a knowledge graph convolution network framework based on the mixed attention coefficients of the user to the relationships and the entity preferences in the RPA knowledge graph, specifically:
determining weights in aggregation according to the preference of the user to the relationship and the entity, and constructing a mixed attention coefficient:
Wherein, Representing the mixed attention coefficient of user u to entity m and entity t,/>Is the relation between entity m and entity t,/>For the representation of user u,/>And/>Respectively the relation/>And representation of entity t,/>And/>Respectively represent the user u pair relation/>And the attention score of entity t, alpha and beta representing two attention scores/>, respectivelyAnd/>Duty ratio of/>Representing a neighborhood set of entity m,/>For the relationship of the mth entity and entity e,/>And/>Respectively represent the user u pair relation/>And an attention score for entity e; the entity in the RPA knowledge graph is an RPA project, and the relationship is the relationship between the RPA project and the RPA project;
Taking the mixed attention coefficient as neighborhood information of a weight aggregation entity m (namely an item m):
Wherein, Neighborhood information aggregation representing user u for entity m (i.e., item m)/>Representation of the representation entity t (i.e. item t)/>Representing a neighborhood set of entity m (i.e., item m);
extending the knowledge graph packing network framework from one layer to ℎ layers, the representation of the final item m is:
Wherein, Representation of item m obtained through h-layer graph convolution neural network learning,/>Representation of item m obtained through h-1 layer graph convolution neural network learning,/>Representing neighborhood information aggregation for item m calculated through a layer h graph convolutional neural network.
Further, a loss function for learning the RPA item representation in the knowledge graph is obtained based on the user representation and the RPA item representationThe expression is as follows:
Wherein, Is a loss function of learning RPA item expression in knowledge graph,/>Is/>And/>BCE loss function of/>Is/>And/>BCE loss function of/>Is a predictive value of the probability of user u interacting with item i,/>Is a true value of the probability of user u interacting with item i,/>Is a predictive value of the probability of user u interacting with item j,/>Is a true value of the probability of user u interacting with item j,/>Representing a positive sampling set, wherein edges exist between user nodes and project nodes in the positive sampling set; /(I)Representing a negative sample set, wherein the user nodes in the negative sample set have no edges with the project nodes.
Further, in step S4, a loss function is combined、/>、/>The final loss function is constructed as follows:
Wherein, As a final loss function,/>、/>、/>Weight parameters of constraint loss terms,/>, respectivelyThe parameter of the person to be learned is represented,Representing a 2-norm.
The invention also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program realizes the steps of the RPA knowledge graph recommendation method when being executed by a processor.
The invention also provides electronic equipment, which comprises a processor and a memory, wherein the processor is connected with the memory, the memory is used for storing a computer program, the computer program comprises computer readable instructions, and the processor is configured to call the computer readable instructions and execute the RPA knowledge graph recommendation method.
The technical scheme provided by the invention has the beneficial effects that:
The invention provides a graphic neural network recommendation method integrating user network and user interaction perception, which not only enhances user representation, but also introduces a higher-order collaborative signal in user interaction into representation learning of a project, thereby improving recommendation accuracy. First, the higher order collaborative signal in user-item interactions is sufficiently captured to enhance the user representation. When learning item representations, the attention function can determine the weight during aggregation according to the user preference of the relationship and the neighbor entity, so that the propagation of the user preference in the knowledge graph can be effectively enhanced, and personalized item representations can be learned.
Drawings
FIG. 1 is a flowchart of an RPA project recommendation method according to an embodiment of the invention;
FIG. 2 is a diagram of a user project bipartite and a user network constructed in accordance with an embodiment of the present invention;
fig. 3 is a block diagram of an electronic device in an exemplary embodiment of the invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be further described with reference to the accompanying drawings.
A flow chart of the RPA project recommending method in the embodiment of the invention is shown in figure 1, and comprises the following steps:
S1, acquiring user project interaction information, constructing a user project bipartite graph, wherein a user and a project are respectively used as a user node and a project node, the edges between the user and the project represent interaction between the user and the project, the user project bipartite graph contains the interaction information of the user, and the hidden high-order cooperative information can effectively help to recommend.
S2, constructing a user network by using a user project bipartite graph based on user representation, constructing a user network G= (A, X) by using the user project bipartite graph, wherein A is an adjacency matrix of user nodes, and represents whether edge connection exists between users,N user representations are learned for two graphs of user items. As shown in FIG. 2, if users of the user project bipartite graph use the same RPA project, the user network constructed by the embodiment of the invention indicates that edges exist among users in the constructed user network. For example, there is a path between user 1 and user 2 and the path does not contain other user nodes, then there is an edge between user 1 and user 2.
S3, acquiring an RPA knowledge graph, wherein the RPA knowledge graph is expressed in the following form: entity-relation-entity, wherein the entity in the RPA knowledge graph is an RPA item, the relation is the relation between the RPA item and the RPA item, and the information aggregation and the propagation are carried out in the RPA knowledge graph by using a knowledge graph convolution network framework based on the mixed attention coefficient of the relation and the entity preference in the RPA knowledge graph by a user, so that the RPA item representation is extracted.
To better focus on user preferences, the ability of the graph convolution network to handle user preference features is enhanced by determining weights at aggregation based on user preferences for entity relationships and entities, constructing a mixed attention coefficient:
Wherein, Representing the mixed attention coefficient of user u to entity m and entity t,/>Is the relation between entity m and entity t,/>For the representation of user u,/>And/>Respectively the relation/>And representation of entity t,/>And/>Respectively represent the user u pair relation/>And the attention score of entity t, alpha and beta representing two attention scores/>, respectivelyAnd/>Duty ratio of/>Representing a neighborhood set of entity m,/>For the relationship of the mth entity and entity e,/>And/>Respectively represent the user u pair relation/>And an attention score for entity e; the entity in the RPA knowledge graph is an RPA item, and the relationship is the relationship between the RPA item and the RPA item.
Taking the mixed attention coefficient as the weight in aggregation, and aggregating the neighborhood information of the item m according to different weights:
Wherein, Neighborhood information aggregation representing user u for entity m (i.e., item m)/>Representation of the representation entity t (i.e. item t)/>Representing a neighborhood set of entity m (i.e., item m).
Extending the knowledge graph packing network framework from one layer to ℎ layers results in a representation of the final entity m, i.e., the representation of item m is:
Wherein, Representation of item m obtained through h-layer graph convolution neural network learning,/>Representation of item m obtained through h-1 layer graph convolution neural network learning,/>Representing neighborhood information aggregation for item m calculated through a layer h graph convolutional neural network.
S4, constructing a recommendation model, wherein the recommendation model comprises the user network and an improved knowledge graph convolution network framework.
The following loss functions need to be used when training the model:
(1) To alleviate the problem of excessive smoothing caused by multi-layer convolution in graph data, a user project bipartite graph loss function is constructed by setting a convergence mode And use/>Instead of the convergence state obtained by the graph convolution model in the user project bipartite graph after multi-hop message transmission, the user representation is learned by using the user project bipartite graph.
The method comprises the following steps:
setting a model convergence state, and when the representation of each user meets the following formula, describing that the model is similar to the convergence state obtained after multi-hop message transmission:
Wherein, Representation of user u, i.e. representation of user u,/>Representing item i, i.e., item i, N (u) represents a historical interaction item for user u,/>Constraint coefficient representing user u and item i,/>Representing user node degree,/>, in user project bipartite graphThe degree of the item node in the user item bipartite graph is represented.
In order to alleviate the problem of excessive smoothing, a negative sampling strategy is adopted in the training process, and constraint loss on the user project bipartite graph is as follows:
Wherein, Representing constraint loss on the user project bipartite graph; /(I)Representing a positive sampling set, wherein the user nodes in the positive sampling set and the project nodes have edges, namely interaction behaviors of the user and the project exist; /(I)Representing a negative sampling set, wherein no edges exist between user nodes and project nodes in the negative sampling set, namely, no interactive behaviors exist between users and projects; /(I)Constraint coefficient representing user u and item i,/>Representing constraint coefficients of user u and item j, σ represents sigmod functions,/>Representation of item j.
In addition to constraint losses, it is also desirable to use BCE losses as the primary optimization objective for learning user representations in user project bipartite graphs as follows:
Wherein, BCE loss representing user project bipartite graph;
The two loss functions use the same sampling set, and the user project bipartite graph loss function is as follows:
Wherein, The user project bipartite graph loss is represented.
(2) And constructing a similarity constraint loss function between user representations, and further relieving the problem of recommending cold start.
The social similarity among users is calculated by adopting cosine similarity as follows:
Wherein, For/>And/>Cosine similarity between/(Representing the ith user node representation, i.e. the ith user representation,/>Representing the j-th user node representation, i.e. the j-th user representation,/>Representing the modulo length of the vector.
The similarity constraint loss between the constructed user representations is as follows:
Wherein, Constraint loss function for similarity between user representations,/>Representing neighbor nodes of the ith user in the user network, and n represents the total number of users.
(3) Optimizing by using a BCE loss and negative sampling strategy, and obtaining a loss function of learning the RPA item representation in the knowledge graph based on the user representation and the RPA item representation as follows:
message aggregation and propagation are carried out in the RPA knowledge graph by using a knowledge graph convolution network framework with improved knowledge, and when the RPA project representation is obtained, dot product is carried out on the RPA project representation and the user representation to predict the probability of interaction between the user and the project:
Wherein, Is a loss function of learning RPA item expression in knowledge graph,/>Is/>And/>BCE loss function of/>Is/>And/>BCE loss function of/>Is a predictive value of the probability of user u interacting with item i,/>Is a true value of the probability of user u interacting with item i,/>Is a predictive value of the probability of user u interacting with item j,/>Is a true value of the probability of user u interacting with item j,/>Representing a positive sampling set, wherein the user nodes in the positive sampling set and the project nodes have edges, namely the user and the project interact; /(I)And representing a negative sampling set, wherein the user nodes in the negative sampling set have no edges with the project nodes, i.e. the user and the project have no interaction.
Bond loss function、/>、/>Constructing a final loss function training recommendation model to obtain an optimal model, deploying the optimal model in a production environment, inputting a user project bipartite graph and an RPA knowledge graph of the production environment to obtain user representation and project representation, solving an inner product of the user representation and the project representation to obtain a prediction result, and recommending the RPA project.
The integral loss function is constructed to optimize the model to obtain a final model M, and the formula is as follows:
Wherein, As a whole loss function,/>As the total loss function of the user project bipartite graph,/>Constraint loss function for similarity between user representations,/>Is a loss function of learning RPA item expression in knowledge graph,/>、/>、/>Weight parameters of constraint loss terms,/>, respectivelyRepresenting a collectible parameter,/>Representing a 2-norm.
The final model M is deployed in a production environment for RPA project recommendation. Inputting a user project bipartite graph (UE) and an RPA Knowledge Graph (KG) of a production environment to obtain user representation and project representation as follows:
Wherein, Representation of n users learned by representation model,/>Representing a representation of m items learned by the model. Obtaining a prediction result by solving an inner product of the user representation and the item representation:
Wherein, ,/>The interaction probability of user i for item j is represented. Based on the predicted results, the top k may be taken as item recommendations for the corresponding user.
In an exemplary embodiment, a computer readable storage medium is further included, where a computer program is stored, and the computer program when executed by a processor implements the steps of the RPA item recommendation method described above.
In an exemplary embodiment, referring to FIG. 3, an electronic device is also included that includes at least one processor, at least one memory, and at least one communication bus.
The processor calls the computer readable instructions stored in the memory through the communication bus to execute the RPA project recommending method.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (9)
1. An RPA project recommendation method is characterized by comprising the following steps:
s1, acquiring user and project information, constructing a user project bipartite graph, and learning user representation by using the user project bipartite graph;
s2, constructing a user network by utilizing a user project bipartite graph based on the user representation;
S3, acquiring an RPA knowledge graph, and extracting an RPA project representation from the RPA knowledge graph by using a knowledge graph convolution network framework based on a mixed attention coefficient of a user on the relationship and entity preference in the RPA knowledge graph;
S4, constructing a recommendation model, wherein the recommendation model comprises a knowledge graph convolution network framework of the user network and a mixed attention coefficient based on the entity relation and entity preference of the user; constructing a user project bipartite graph loss function by setting model convergence mode ; Based on the user network, constructing a similarity constraint loss function/>, between user representations; Obtaining a loss function/>, based on the user representation and the RPA item representation, of learning the RPA item representation in the knowledge graph; Binding loss function/>、/>、/>Constructing a final loss function training recommendation model to obtain a final recommendation model;
And S5, deploying the final recommendation model in a production environment, inputting a user project bipartite graph and an RPA knowledge graph of the production environment, obtaining user representation and RPA project representation of the production environment, and obtaining a prediction result by solving an inner product of the user representation and the RPA project representation of the production environment, wherein the prediction result is used for recommending the RPA project.
2. The RPA project recommendation method according to claim 1, wherein in step S2, based on the user representation, constructing the user network using the user project bipartite graph is specifically:
constructing a user network G= (A, X) by using a user project bipartite graph, wherein A is an adjacency matrix of user nodes in the user project bipartite graph, and represents whether edge connection exists between users, N user representations are learned for two graphs of user items.
3. The method for recommending an RPA project according to claim 1, wherein the user project bipartite graph loss function is constructed by setting a model convergence modeThe method specifically comprises the following steps:
Setting a model convergence state:
Wherein, Representation of representation user u,/>Representation of representation item i, N (u) represents a historical interaction item for user u,/>Constraint coefficient representing user u and item i,/>Representing user node degree,/>, in user project bipartite graphRepresenting the node degree of the item in the two-part diagram of the user item;
constraint loss on the user project bipartite graph is as follows:
Wherein, Representing constraint loss on the user project bipartite graph; /(I)Representing a positive sampling set, wherein edges exist between user nodes and project nodes in the positive sampling set; /(I)Representing a negative sampling set, wherein no edges exist between user nodes and project nodes in the negative sampling set; /(I)Representing constraint coefficients of user u and item j, σ represents sigmod functions,/>A representation of a representation item j;
using BCE loss as the primary optimization objective for learning user representations in a user project bipartite graph, the BCE loss for a user project bipartite graph is as follows:
Wherein, BCE loss representing user project bipartite graph;
The user project bipartite graph loss function is as follows:
Wherein, A graph loss function for the user project bipartite.
4. The RPA project recommendation method according to claim 1, wherein a similarity constraint loss function between user representations is constructed based on the user networkThe method specifically comprises the following steps:
the social similarity among users is calculated by adopting cosine similarity as follows:
Wherein, For/>And/>Cosine similarity between/(Representing the ith user representation,/>Representing the j-th user representation,/>Representing the modular length of the vector;
the similarity constraint loss between the constructed user representations is as follows:
Wherein, Constraint loss function for similarity between user representations,/>Representing neighbor nodes of the ith user in the user network, and n represents the total number of users.
5. The RPA item recommendation method according to claim 1, wherein in step S3, the RPA item representation is extracted from the RPA knowledge graph using a knowledge graph convolution network framework based on a mixed attention coefficient of a user to a relationship and an entity preference in the RPA knowledge graph, specifically:
determining weights in aggregation according to the preference of the user to the relationship and the entity, and constructing a mixed attention coefficient:
Wherein, Representing the mixed attention coefficient of user u to entity m and entity t,/>For the relationship of entity m and entity t,For the representation of user u,/>And/>Respectively the relation/>And representation of entity t,/>And/>Respectively represent the user u pair relationAnd the attention score of entity t, alpha and beta representing two attention scores/>, respectivelyAnd/>Duty ratio of/>Representing a neighborhood set of entity m,/>For the relationship of the mth entity and entity e,/>And/>Respectively represent the user u pair relation/>And an attention score for entity e; the entity in the RPA knowledge graph is an RPA project, and the relationship is the relationship between the RPA project and the RPA project;
Aggregating neighborhood information of item m by taking the mixed attention coefficient as weight:
Wherein, Neighborhood information aggregation representing user u versus entity m,/>Representation of representation entity t,/>Representing a neighborhood set of entity m;
extending the knowledge graph packing network framework from one layer to the h layer, results in a representation of the final entity m, i.e. the representation of item m is:
Wherein, Representation of item m obtained through h-layer graph convolution neural network learning,/>Representation of item m obtained through h-1 layer graph convolution neural network learning,/>Representing neighborhood information aggregation for item m calculated through a layer h graph convolutional neural network.
6. The method of claim 5, wherein the learning of the RPA item representation in the knowledge graph is based on a user representation and the RPA item representation to obtain a loss functionThe expression is as follows:
Wherein, Is a loss function of learning RPA item expression in knowledge graph,/>Is/>And/>BCE loss function of/>Is/>And/>BCE loss function of/>Is a predictive value of the probability of user u interacting with item i,/>Is a true value of the probability of user u interacting with item i,/>Is a predictive value of the probability of user u interacting with item j,/>Is a true value of the probability of user u interacting with item j,/>Representing a positive sampling set, wherein edges exist between user nodes and project nodes in the positive sampling set; /(I)Representing a negative sample set, wherein the user nodes in the negative sample set have no edges with the project nodes.
7. The method according to claim 1, wherein in step S4, a loss function is combined、/>、/>The final loss function is constructed as follows:
Wherein, As a final loss function,/>、/>、/>Weight parameters of constraint loss terms,/>, respectivelyRepresenting a collectible parameter,/>Representing a 2-norm.
8. A computer-readable storage medium storing a computer program, characterized in that: the computer program implementing the steps of the method according to any of claims 1-7 when executed by a processor.
9. An electronic device comprising a processor and a memory, the processor being interconnected with the memory, wherein the memory is configured to store a computer program comprising computer readable instructions, the processor being configured to invoke the computer readable instructions to perform the method of any of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410383490.1A CN117972231B (en) | 2024-04-01 | 2024-04-01 | RPA project recommendation method, storage medium and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410383490.1A CN117972231B (en) | 2024-04-01 | 2024-04-01 | RPA project recommendation method, storage medium and electronic equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117972231A true CN117972231A (en) | 2024-05-03 |
CN117972231B CN117972231B (en) | 2024-06-11 |
Family
ID=90863514
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410383490.1A Active CN117972231B (en) | 2024-04-01 | 2024-04-01 | RPA project recommendation method, storage medium and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117972231B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114637923A (en) * | 2022-05-19 | 2022-06-17 | 之江实验室 | Data information recommendation method and device based on hierarchical attention-graph neural network |
CN114880559A (en) * | 2022-04-28 | 2022-08-09 | 北方民族大学 | User-project fused neighbor entity representation recommendation method |
CN115905691A (en) * | 2022-11-11 | 2023-04-04 | 云南师范大学 | Preference perception recommendation method based on deep reinforcement learning |
CN116050523A (en) * | 2023-01-30 | 2023-05-02 | 天津大学 | Attention-directed enhanced common sense reasoning framework based on mixed knowledge graph |
CN116091152A (en) * | 2022-10-07 | 2023-05-09 | 西安工业大学 | Recommendation method and system based on multi-level comparison learning and multi-mode knowledge graph |
CN116127177A (en) * | 2022-10-26 | 2023-05-16 | 北京交通大学 | Recommendation method for embedding negative sampling optimization by utilizing knowledge graph |
CN116340595A (en) * | 2023-02-27 | 2023-06-27 | 之江实验室 | Fairness recommendation algorithm based on knowledge graph |
CN117520522A (en) * | 2023-12-29 | 2024-02-06 | 华云天下(南京)科技有限公司 | Intelligent dialogue method and device based on combination of RPA and AI and electronic equipment |
-
2024
- 2024-04-01 CN CN202410383490.1A patent/CN117972231B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114880559A (en) * | 2022-04-28 | 2022-08-09 | 北方民族大学 | User-project fused neighbor entity representation recommendation method |
CN114637923A (en) * | 2022-05-19 | 2022-06-17 | 之江实验室 | Data information recommendation method and device based on hierarchical attention-graph neural network |
CN116091152A (en) * | 2022-10-07 | 2023-05-09 | 西安工业大学 | Recommendation method and system based on multi-level comparison learning and multi-mode knowledge graph |
CN116127177A (en) * | 2022-10-26 | 2023-05-16 | 北京交通大学 | Recommendation method for embedding negative sampling optimization by utilizing knowledge graph |
CN115905691A (en) * | 2022-11-11 | 2023-04-04 | 云南师范大学 | Preference perception recommendation method based on deep reinforcement learning |
CN116050523A (en) * | 2023-01-30 | 2023-05-02 | 天津大学 | Attention-directed enhanced common sense reasoning framework based on mixed knowledge graph |
CN116340595A (en) * | 2023-02-27 | 2023-06-27 | 之江实验室 | Fairness recommendation algorithm based on knowledge graph |
CN117520522A (en) * | 2023-12-29 | 2024-02-06 | 华云天下(南京)科技有限公司 | Intelligent dialogue method and device based on combination of RPA and AI and electronic equipment |
Non-Patent Citations (2)
Title |
---|
张素琪等: "基于双用户视角与知识图谱注意力网络的推荐模型", 现代计算机, 15 June 2020 (2020-06-15) * |
陈平华;朱禹;: "融合知识图谱表示学习和矩阵分解的推荐算法", 计算机工程与设计, no. 10, 16 October 2018 (2018-10-16) * |
Also Published As
Publication number | Publication date |
---|---|
CN117972231B (en) | 2024-06-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102208086B (en) | Field-oriented personalized intelligent recommendation system and implementation method | |
CN112364242B (en) | Graph convolution recommendation system for context awareness | |
CN111881345B (en) | Neural collaborative filtering service quality prediction method based on position context awareness | |
CN115203550A (en) | Social recommendation method and system for enhancing neighbor relation | |
CN113590976A (en) | Recommendation method of space self-adaptive graph convolution network | |
CN115760279A (en) | Knowledge graph and multi-head attention-based dual-target cross-domain recommendation method and system | |
CN115618098A (en) | Cold-chain logistics recommendation method and device based on knowledge enhancement and hole convolution | |
CN117216281A (en) | Knowledge graph-based user interest diffusion recommendation method and system | |
CN116680469A (en) | Sequence recommendation algorithm based on dynamic graph neural network | |
CN111626827A (en) | Method, device, equipment and medium for recommending articles based on sequence recommendation model | |
CN115631008B (en) | Commodity recommendation method, device, equipment and medium | |
CN113688974B (en) | Mobile application recommendation method based on lightweight graph convolutional network | |
CN114417172A (en) | Deep interest evolution recommendation method, device, equipment and storage medium | |
CN117972231B (en) | RPA project recommendation method, storage medium and electronic equipment | |
CN115545833A (en) | Recommendation method and system based on user social information | |
CN111598093A (en) | Method, device, equipment and medium for generating structured information of characters in picture | |
CN116467466A (en) | Knowledge graph-based code recommendation method, device, equipment and medium | |
Horling et al. | Quantitative organizational models for large-scale agent systems | |
CN112765474B (en) | Recommendation method and system based on depth collaborative filtering | |
CN115391555A (en) | User-perceived knowledge map recommendation system and method | |
CN114692012A (en) | Electronic government affair recommendation method based on Bert neural collaborative filtering | |
CN112035567A (en) | Data processing method and device and computer readable storage medium | |
CN117788122B (en) | Goods recommendation method based on heterogeneous graph neural network | |
Chen et al. | Social recommendation using quantified social tie strength | |
CN118013060B (en) | Data processing method, device, equipment, storage medium and product |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |