CN115952298A

CN115952298A - Supplier performance risk analysis method and related equipment

Info

Publication number: CN115952298A
Application number: CN202211634589.1A
Authority: CN
Inventors: 毕艳冰; 李向阳; 姜凯华; 田青; 师择鹏
Original assignee: Beijing Sgitg Accenture Information Technology Co ltd
Current assignee: Beijing Sgitg Accenture Information Technology Co ltd
Priority date: 2022-12-19
Filing date: 2022-12-19
Publication date: 2023-04-11

Abstract

The application provides a supplier fulfillment risk analysis method and related equipment. The method comprises the following steps: constructing an ontology according to the acquired supplier parameters; extracting the acquired supplier data based on a pre-trained entity extraction model to obtain entity knowledge; extracting the text data set by using a relation extraction framework to obtain an entity relation; extracting the statement set by using an event extraction model to obtain event knowledge; performing knowledge fusion based on the ontology, the entity knowledge, the entity relationship and the event knowledge to obtain a knowledge graph; and fusing with real-time major events of the suppliers on the basis of a knowledge graph so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with performing risks. According to the embodiment of the application, the supplier performance risk knowledge graph is subjected to knowledge fusion with the major time sequence and the causal events, so that the holographic evaluation on the supplier is realized in real time, and the performance risk of the supplier is predicted.

Description

Supplier performance risk analysis method and related equipment

Technical Field

The application relates to the technical field of knowledge graphs, in particular to a supplier performance risk analysis method and related equipment.

Background

In the professional auditing process of engineering, materials and the like, the service condition of a supplier needs to be paid attention mainly, including the related information of the qualification, the service range, the operation performance, the equity relationship, the industry dynamics and the like of the supplier, and the risk that the supplier cannot fulfill the contract is pre-warned. However, with the advent of the big data information age, auditors have spent a great deal of time mining valuable data from a vast amount of disparate provider enterprise activity data. How to utilize big data technology, the intelligent decision-making of supplementary audit personnel reduces the human cost, is the research content that urgent need solved.

The problem of 'old and difficult' in the material performance work is that the actual production and operation conditions of the enterprises of the suppliers are difficult to accurately master. At present, the unified aggregation and service of the full data based on a supplier as a base point are not available, missing items are difficult to avoid if the aggregation is carried out by a manual method, and the data combing efficiency of multi-layer association is low. The method has the advantages that the large data technology is urgently needed to realize holographic multi-dimensional real-time evaluation on suppliers, and auxiliary support is provided for business scenes of tender procurement and quality guarantee performance.

Disclosure of Invention

In view of the above, an object of the present application is to provide a supplier fulfillment risk analysis method and related apparatus.

In view of the above, the present application provides a supplier fulfillment risk analysis method, including:

according to the obtained supplier parameters, an ontology in a supplier knowledge graph is constructed;

extracting the obtained supplier data based on a pre-trained entity extraction model to obtain supplier entity knowledge;

extracting a text data set of a supplier by using a preset relation extraction frame to obtain a supplier entity relation;

extracting the statement set of the supplier by using a pre-trained event extraction model to obtain supplier event knowledge;

performing knowledge fusion based on the ontology, the supplier entity knowledge, the supplier entity relationship and the supplier event knowledge to obtain the supplier knowledge graph;

and fusing the real-time important events of the suppliers based on the knowledge graph of the suppliers so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with the performing risk.

In a possible implementation manner, the constructing an ontology in a provider knowledge graph according to the obtained provider parameters includes:

defining the application field and knowledge range of the ontology;

abstract description is carried out on concepts and entities in the application fields and the knowledge range;

defining attributes and attribute values of entities in the ontology;

adding rule constraints to the ontology to build the ontology in the provider knowledge graph.

In one possible implementation, the training process of the pre-trained entity extraction model includes:

acquiring an original supplier data set;

extracting data in the original supplier data set based on a pre-compiled regular expression to obtain a data set of a hit rule;

the data set of the hit rule comprises a training data set and a prediction data set; training the entity extraction model to be trained by using the training data set;

and stopping training to obtain the pre-trained entity extraction model in response to the accuracy rate obtained by inputting the prediction data set into the entity extraction model to be trained being higher than a preset first threshold value.

In a possible implementation manner, the extracting a text data set of a provider by using a preset relationship extraction framework to obtain a provider entity relationship includes:

matching the obtained original text data set based on a preset first seed entity to obtain a first entity relationship example set;

extracting a template corresponding to the first entity relationship instance set to obtain a template library;

extracting a template from the template library by using a single clustering algorithm to obtain an extraction relation template;

extracting the original text data set based on the extraction relation template to obtain a second entity relation instance set;

and in response to the size of the second entity relationship instance set reaching a preset second threshold, taking the second entity relationship instance set as the provider entity relationship.

In a possible implementation manner, the extracting the statement set of the provider by using the pre-trained event extraction model to obtain provider event knowledge includes:

inputting the supplier's statement set into the pre-trained event extraction model;

in the pre-trained event extraction model, expanding each sentence in the sentence set of the supplier to a fixed preset length to obtain a sentence set with a preset length; coding each word in each sentence in the preset length sentence set to obtain a word vector set with fixed dimensions; and capturing context information of sentences in the sentence set of the supplier based on the word vector set with the fixed dimension, and performing global optimization on the context information by utilizing a fixed rate coefficient layer to obtain the supplier event knowledge.

In one possible implementation, the knowledge fusion includes: entity knowledge fusion and event knowledge fusion;

the obtaining the provider knowledge graph by performing knowledge fusion based on the ontology, the provider entity knowledge, the provider entity relationship, and the provider event knowledge, includes:

based on the ontology, the vendor entity knowledge, and the vendor entity relationships, performing the entity knowledge fusion using a framework matching technique, an entity alignment technique, and/or a conflict detection and resolution technique;

carrying out named entity extraction based on the entity, the supplier entity relationship and the supplier event knowledge to obtain a supplier event knowledge entity so as to carry out event knowledge fusion;

and constructing and obtaining the provider knowledge graph based on the entity knowledge fusion and the event knowledge fusion.

In one possible implementation, the method further includes:

obtaining a heterogeneous provider knowledge graph that is heterogeneous to the provider knowledge graph;

encoding the entities and the relations in the knowledge graph of the heterogeneous supplier to obtain a low-dimensional semantic micro space;

and based on a preset alignment entity seed set, carrying out entity alignment on the entities between the supplier knowledge graph and the heterogeneous supplier knowledge graph according to the supplier knowledge graph, the internal structure of the heterogeneous supplier knowledge graph and the semantic distance between the entities in the low-dimensional semantic micro space.

Based on the same inventive concept, the embodiment of the present application further provides a supplier fulfillment risk analysis device, including:

a building module configured to build an ontology in a supplier knowledge graph according to the obtained supplier parameters;

the extraction module is configured to extract the acquired supplier data based on a pre-trained entity extraction model to obtain supplier entity knowledge;

the extraction module is configured to extract a text data set of a supplier by using a preset relation extraction framework to obtain a supplier entity relation;

the extraction module is also configured to extract the statement set of the supplier by using a pre-trained event extraction model to obtain supplier event knowledge;

a fusion module configured to perform knowledge fusion based on the ontology, the supplier entity knowledge, the supplier entity relationship, and the supplier event knowledge to obtain the supplier knowledge graph;

the fusion module is also configured to fuse real-time major events of the suppliers based on the knowledge graph of the suppliers so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with the performing risk.

Based on the same inventive concept, embodiments of the present application further provide an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the method for analyzing provider performance risk as described in any one of the above embodiments is implemented.

Based on the same inventive concept, embodiments of the present application further provide a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute any one of the above-mentioned supplier fulfillment risk analysis methods.

From the above, according to the supplier fulfillment risk analysis method and the related equipment provided by the application, an ontology in a supplier knowledge graph is constructed according to the acquired supplier parameters; extracting the acquired supplier data based on a pre-trained entity extraction model to obtain supplier entity knowledge; extracting a text data set of a supplier by using a preset relation extraction frame to obtain a supplier entity relation; extracting the statement set of the supplier by using a pre-trained event extraction model to obtain supplier event knowledge; performing knowledge fusion based on the ontology, the supplier entity knowledge, the supplier entity relationship and the supplier event knowledge to obtain the supplier knowledge graph; and fusing the real-time important events of the suppliers based on the knowledge graph of the suppliers so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with the performing risk. By utilizing various technologies of entity extraction, relation extraction, event extraction, knowledge fusion and the like of natural language processing, internal and external data of business license information, stockholder information, main personnel of an enterprise, external investment, judicial cases, patents, contract fulfillment and the like of a supplier unit are collected, a supplier fulfillment risk analysis library is established, and a knowledge graph facing the supplier fulfillment risk audit is formed. And extracting the major time sequence and causal events of the supplier, and performing knowledge fusion with the knowledge map of the auditing supplier to form clear evaluation and analysis data of the supplier, so as to realize real-time holographic evaluation of the supplier and predict the performance risk of the supplier.

Drawings

In order to more clearly illustrate the technical solutions in the present application or the related art, the drawings needed to be used in the description of the embodiments or the related art will be briefly introduced below, and it is obvious that the drawings in the following description are only embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.

FIG. 1 is a flow chart of a supplier fulfillment risk analysis method according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a technical route of an entity extraction model according to an embodiment of the present application;

FIG. 3 is a schematic diagram of a relationship extraction framework according to an embodiment of the present application;

FIG. 4 is a schematic diagram of an event extraction model according to an embodiment of the present application;

FIG. 5 is a diagram illustrating an example of a fusion of entity associations according to an embodiment of the present application;

FIG. 6 is a schematic structural diagram of an apparatus for analyzing a supplier fulfillment risk according to an embodiment of the present application;

fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is further described in detail below with reference to the accompanying drawings in combination with specific embodiments.

It should be noted that technical terms or scientific terms used in the embodiments of the present application should have a general meaning as understood by those having ordinary skill in the art to which the present application belongs, unless otherwise defined. The use of "first," "second," and similar terms in the embodiments of the present application do not denote any order, quantity, or importance, but rather the terms are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that the element or item preceding the word comprises the element or item listed after the word and its equivalent, but does not exclude other elements or items. The terms "connected" or "coupled" and the like are not restricted to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", and the like are used merely to indicate relative positional relationships, and when the absolute position of the object being described is changed, the relative positional relationships may also be changed accordingly.

As described in the background section, in the related art, in the professional auditing process of engineering, materials and the like, the service condition of the supplier needs to be focused on, including the related information such as the qualification, business scope, operation performance, equity relationship, industry dynamics and the like of the supplier, and the risk that the supplier cannot fulfill the contract is pre-warned. However, with the advent of the big data information age, auditors have spent a great deal of time mining valuable data from a vast amount of disparate provider enterprise activity data. How to utilize big data technology, the intelligent decision-making of auxiliary audit personnel reduces the human cost, is the research content that urgent needs to solve.

The problem that the actual production and operation conditions of a supplier enterprise are difficult to accurately master is the 'old and difficult' problem in the material performance work. At present, the unified aggregation and service of the full data based on a supplier as a base point are not available, missing items are difficult to avoid if the aggregation is carried out by a manual method, and the data combing efficiency of multi-layer association is low. The method has the advantages that the large data technology is urgently needed to realize holographic multi-dimensional real-time evaluation on suppliers, and auxiliary support is provided for business scenes of tender procurement and quality guarantee performance.

In view of the above, the embodiment of the present application provides a supplier fulfillment risk analysis method, which constructs an ontology in a supplier knowledge graph according to acquired supplier parameters; extracting the acquired supplier data based on a pre-trained entity extraction model to obtain supplier entity knowledge; extracting a text data set of a supplier by using a preset relation extraction frame to obtain a supplier entity relation; extracting the statement set of the supplier by using a pre-trained event extraction model to obtain supplier event knowledge; performing knowledge fusion based on the ontology, the supplier entity knowledge, the supplier entity relationship and the supplier event knowledge to obtain the supplier knowledge graph; and fusing the real-time important events of the suppliers based on the knowledge graph of the suppliers so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with the performing risk. By utilizing various technologies of entity extraction, relation extraction, event extraction, knowledge fusion and the like of natural language processing, internal and external data such as business license information, stockholder information, enterprise major personnel, external investment, judicial cases, patents, contract fulfillment and the like of a supplier unit are collected, a supplier fulfillment risk analysis library is established, and a knowledge map facing the supplier fulfillment risk audit is formed. And extracting the major time sequence and causal events of the supplier, and performing knowledge fusion with the knowledge map of the auditing supplier to form clear evaluation and analysis data of the supplier, so as to realize real-time holographic evaluation of the supplier and predict the performance risk of the supplier.

Hereinafter, the technical means of the embodiments of the present application will be described in detail by specific examples.

Referring to fig. 1, a supplier fulfillment risk analysis method according to an embodiment of the present application includes the following steps:

step S101, an ontology in a knowledge graph of a supplier is constructed according to the acquired supplier parameters;

step S102, extracting the acquired supplier data based on a pre-trained entity extraction model to obtain supplier entity knowledge;

step S103, extracting a text data set of a supplier by using a preset relation extraction framework to obtain a supplier entity relation;

step S104, extracting the statement set of the supplier by using a pre-trained event extraction model to obtain the supplier event knowledge;

step S105, based on the ontology, the supplier entity knowledge, the supplier entity relationship and the supplier event knowledge, performing knowledge fusion to obtain the supplier knowledge graph;

and step S106, fusing the real-time major events of the suppliers based on the knowledge graph of the suppliers so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with performing risks.

Aiming at the step S101, the ontology is used for describing concepts and related knowledge in a specific field, the knowledge which is highly summarized and refined is stored in the ontology, the ontology generally has high quality, the concept ontology base is generally used for management, whether the definition of the ontology base is perfect and accurate directly influences the quality of the constructed knowledge graph, and the ontology base can realize the unified specification and management of entities, relationships among the entities and attributes of the entities in the knowledge graph and is beneficial to improving the quality of the knowledge graph. Since the subject is a knowledge graph oriented to the provider performance risk field, the concept is relatively clear, and the scope is relatively fixed, the concept ontology base is constructed in a top-down manner. The following mainly introduces the construction method and process of the ontology library in the field of business fulfillment risk analysis of the supplier.

In a knowledge graph, an ontology belongs to a concept layer, and the ontology is a description for abstracting various concepts and knowledge in the real world. The entity and the relation thereof in the knowledge map can be restrained and normalized by constructing the ontology, so that the management of the knowledge on the basis of a unified ontology base is facilitated, and the improvement of the quality of knowledge in the knowledge base is facilitated.

In the process of constructing the ontology, all concepts and knowledge in the field should be included as much as possible, which is beneficial to expanding the application range of the ontology, but in the case of limited resources, it is impractical and risky to pursue the coverage range of the ontology excessively. Therefore, before an ontology is constructed, it is necessary to clarify the application range of the ontology and to investigate and analyze concepts and knowledge in the field in detail. In addition, a set of standardized flows can be used as reference in the body construction process, the quality of the body construction is guaranteed, actual requirements are met, and through investigation and analysis, the following flows can be adopted to complete the construction of the body in the audit field: defining the application field and knowledge range of the ontology; through investigation and analysis, abstract description is carried out on concepts and entities in the field range; defining attributes and attribute values owned by entities in the ontology; add some additional rule constraints to the ontology; and completing the construction of the ontology instance.

For step S102, the obtained supplier data includes structured data and unstructured data.

Structured data refers to a database or table data and the like, generally, the structured data is mostly relational data with prior experience, which is arranged by enterprises, generally, the data quality is reliable, and the data can be directly extracted by adopting a template matching and regularization rule extraction method.

Aiming at the problem of extracting an audit special entity, the traditional technical scheme is that a rule extraction method is utilized, namely, an expert writes rules such as a regular expression aiming at context information of specific content in a webpage, and extracts accurate key information from the rules. However, this method requires the development of rule compilation, maintenance and testing for all sites, is heavy in workload and prone to errors, and is difficult to adapt to the requirement of entity extraction in the cross-service-domain auditing field.

Therefore, the embodiment of the application adopts a composite technical route combining the sequence labeling and rule extraction methods. The classical entity extraction method based on the sequence labeling algorithm has the advantage of strong adaptability, can automatically summarize the mode through the algorithm (form an extraction model), and has strong extraction capability for extracting key information which has no obvious mode and is difficult to observe a specific rule manually. However, this method also has its inherent disadvantages, including: a certain amount of already labeled corpora are required to be imported, and the labeling work of the corpora needs to be manually written. The more the key information to be extracted lacks of modes, the more accurate the extraction result is required to be, and the more linguistic data need to be introduced; in addition, as a machine learning algorithm, the sequence annotation is not stable enough, the execution process is "black box" (unlike the regular judgment that can go back), the accuracy is not determined by the algorithm, but mainly depends on whether the annotated corpus used for training is consistent with the target test corpus, so the "extraction model" constructed by the method is often difficult to judge whether the requirement of the business on the extraction accuracy can be met.

Therefore, by combining actual requirements and data characteristics, a sequence labeling algorithm is adopted, and meanwhile a 'rule method' is introduced into the technical route of the application, so that initial driving corpora are provided for the construction of a sequence labeling algorithm model, and the whole process is subjected to standardized engineering definition, so that the problems that more manual labeling corpora are prepared in advance, the extraction effect is unstable, and the accuracy is difficult to predict are solved, and the universality and the effect stability of the technology are improved.

Referring to fig. 2, a technical route diagram of an entity extraction model according to an embodiment of the present application is shown.

Specifically, firstly, the characteristics of high accuracy and small matching range of 'regular expression extraction' are utilized to compile a small amount of rules, so that a small amount of accurate extraction results are matched from a large amount of linguistic data and are used as the import of the subsequent process.

Furthermore, a certain proportion (80%) of the extraction result obtained in the process is cut out and used as a training corpus led into an automatic sequence labeling method to replace a manual labeling process.

Furthermore, an extraction model is constructed by utilizing the training corpus and combining an open-source automatic sequence labeling class algorithm.

Further, the result of the step 3 is utilized to automatically extract the residual linguistic data (20%) cut out in the step 2, and the extraction result is automatically judged; if the accuracy of the automatic judgment of the model does not meet the service requirement, rewriting more regular expressions to form more 'marking corpora' to be used as model training import; and if the accuracy of the automatic judgment of the model meets the service requirement, stopping the process, and taking the model as a final model deployment application of text extraction.

Compared with the prior art, the technical route uses the 'rule judgment' based on a small amount of rules to replace 'manual labeling' to obtain the initial standard corpus, so that the initial manual input is greatly reduced; the extraction result of the sequence labeling model is automatically retested to ensure that the accuracy of the model meets the service requirement; and the sequence label is used as an executor for extracting the final text, so that the applicability of the model is ensured (the method is not limited to whether the information to be extracted has a strict template or not, and the extraction range is far higher than that of a method based on a 'regular expression'). The entire process is incrementally iterative. If the extraction effect of the model obtained by the sequence labeling training is not ideal (the accuracy cannot meet the requirement), only a small number of regular expressions need to be additionally written, the same process cycle is executed, the extraction effect of the model can be effectively improved, and the rules written in the early stage cannot be abandoned.

In step S103, in the provider entity relationship extraction, in order to avoid consuming a large amount of manpower and material resources for manual labeling in the entity relationship extraction, a semi-supervised learning method may be used to perform entity extraction, and semi-supervised learning is one of the commonly used methods. According to the method, a Bootstrapping method is adopted to conduct semi-supervised supplier entity relation extraction.

Bootstrapping, also known as self-expanding technology, is a machine learning technology that is widely used in various fields. A small amount of labeled information is used as initialization information, and unlabeled information with common characteristics with the labeled information is added in an iterative mode until the required information scale is reached. Therefore, semi-supervised learning is one of the more common methods, which needs to extract a part of seed sets from texts as the basis of self-help expansion, the most common method for selecting seed sets is a random selection method, then entity pairs meeting conditions are manually marked out as seed sets, the method can select entity pairs with certain representativeness in corpus sets as seeds, match texts and extract templates of the texts according to the entity pairs in the seed sets, calculate the similarity between the templates corresponding to the texts in the remaining corpus and the templates in a template library, select N templates with the highest reliability according to a certain strategy and add the N templates into the template library, then expand the current seed set based on the template library matching entity pairs, and continue the next iteration until the end.

The main defect of semi-supervised learning is that semantic drift phenomenon exists, namely, along with the iteration, the newly extracted semantic relationship has a warp and the semantic relationship expressed in the original seed data set has a deviation. The main reason for semantic drift is that extended text instances do not represent relationship semantics correctly. These wrong text examples would extract the wrong pattern representation method, resulting in a reduction of the extraction effect. The traditional semi-supervised learning adopts TF-IDF as the basis of a calculation template, and semantic drift is easily caused. Therefore, the word vector based on the pre-training language model is used as the technical route of the template in the Bootstrapping process, the relation example in the text can be found out more accurately, and the accuracy of text relation extraction is improved.

Referring to fig. 3, a schematic diagram of a relationship extraction framework according to an embodiment of the present application is shown.

Firstly, a text set is scanned, and a corresponding text instance is found according to the seed entity pair. Specifically, all text instances are scanned and if two entities in a pair of seed entities appear in one sentence at the same time, the text is extracted and represented in the form of a five-tuple: < BEF, e1, BET, e2, AFT >. Where BEF is the context before the first entity appears, BET is the context between the two entities, AFT is the context after the second entity, e1 is the first entity, and e2 is the second entity. In the BET section, a relational model based on shallow heuristics is employed. This method limits the contents of the BET segment to words of verb nature, such as verbs, verb phrases, and the like. If there are no words of verb nature between two entities, all words between the two entities are extracted to constitute a BET. Next, each content segment (BEF, BET, AFT) is converted into an independent vector by a pre-trained language model (e.g. BERT), and a vector representation of the entire content segment is obtained by combining the word vectors of each word.

And further, generating a template library for the template corresponding to the text instance which is further matched in the extraction. The template is then extracted using a single clustering algorithm. Corresponding to the text representation method, the clustering result is also composed of three segments of vectors, and the specific process is as follows: firstly, inputting a text instance list, and distributing a first text instance to a null cluster; and then traversing all the text instances in the list, calculating the representation vector of the text instance and the similarity of each cluster for each text instance, and dividing the text instance into a first class cluster with the confidence level higher than a certain threshold value.

Further, after the extraction template is generated, the next task is to extract more text instances in conjunction with the template. First, all documents need to be rescanned, and the text is expressed in the form of (BEF, BET, AFT); then, the similarity between all the existing templates is calculated. When the similarity between the text and the category is equal to or higher than a threshold value, the text may be considered as a candidate text instance.

For step S104, the conventional recurrent neural network cannot learn a long-term context dependency relationship, which results in poor performance in the sequence labeling task. The LSTM network intervenes in the information transmission process in the network by introducing a memory unit and a door mechanism, can better remember the characteristics of the context, has good effect, and is widely applied to various sequence labeling tasks, such as word segmentation, part of speech labeling, named entity identification and the like. The Bi-LSTM model is composed of two LSTM network structures, namely a forward LSTM and a backward LSTM, can capture the context dependency relationship of a text sequence from front to back and from back to front at the same time, and has better effect on a sequence marking task compared with a single-layer LSTM network.

Therefore, the Bi-LSTM + CRF network structure diagram is adopted, the model structure is similar to that of the Bi-LSTM, the last Softmax output layer is changed into a conditional random field, and the model mainly comprises an input layer, a word embedding layer, a Bi-LSTM layer and a CRF layer.

Referring to fig. 4, a schematic diagram of an event extraction model according to an embodiment of the present application is shown.

In the present embodiment, in the input layer, since the input layer of the neural network is usually a fixed dimension, and the text is usually different in length, it is first necessary to extend each sentence in the input to a fixed length, and to fill up the sentences whose length is smaller than the fixed length with "unknown".

In the Embedding layer, each word in a sentence is encoded into a word vector of a fixed dimension, typically 200 dimensions, by one-hot. Thus, a sequence of words can be expressed as:

w＝{w ₁ ,…w _t ,w _t+1 …w _n }

wherein, w _t A word vector representing one d dimension represents the t-th word in the sentence and n represents the length of the sentence.

In the Bi-LSTM coding layer, a Bi-LSTM model is composed of two LSTM network structures, namely a forward LSTM and a backward LSTM, and context information of a sentence can be captured and high-dimensional features in the sentence can be extracted by means of a gate control structure of the LSTM.

In the CRF output layer, the last Softmax layer of the original model is changed to the CRF layer in this embodiment. Therefore, high-dimensional features in data can be extracted by using the Bi-LSTM layer, and meanwhile, the characteristics of CRF global optimization are combined, so that the model learns strong constraint conditions at a plurality of sentence levels, the accuracy of event extraction is improved, and the defect of local optimization of the Bi-LSTM is overcome.

Further, in step S105, the entity knowledge fusion technique based on the vectorization model mainly includes a frame matching technique, an entity alignment technique, a conflict detection and resolution technique, and the like. Through carrying out knowledge fusion on the extracted audit knowledge, the accuracy and consistency of the knowledge are improved, and a foundation is laid for constructing a high-quality audit field knowledge map.

With the deep progress of the construction of modern intelligent supply chains, a better foundation of a related knowledge system in the field of provider performance risk analysis is formed, the domain knowledge is modeled and expressed on the cognitive and semantic levels, commonly recognized words in the domain are determined, the entities are described through the relationship between concepts, and the common understanding of the domain knowledge is provided. However, the heterogeneity of the knowledge system is caused by the self-dispersity of the knowledge system, namely, different knowledge systems are difficult to combine. Framework matching can solve this problem and is an important component of knowledge fusion.

The framework matching can be classified into element-level matching and structure-level matching according to the used technology. And the element-level matching independently judges whether the elements in the two systems are matched or not, and does not consider the matching condition of other elements. The structure level matching does not take each element as an isolated resource, but utilizes the structure of the resource map to consider the influence of the matching condition of other related elements in the element matching process.

The most basic method of element-level matching is based on a character string matching technology, and matching methods such as prefix distance, suffix distance, edit distance and n-gram distance are adopted. The basic idea of structure level matching is: similar concepts have similar structures, and the main techniques include: graph-based techniques, classification-based techniques, and statistical analysis-based techniques.

Entity alignment, also referred to as entity matching, is the process of determining whether two entities in the same or different repositories represent the same physical object. Entity alignment can be divided into two different algorithms, paired entity alignment and collaborative entity alignment. The paired entity alignment means that whether two entities correspond to the same physical object is independently judged, and the alignment degree of the two entities is judged by matching characteristics such as entity attributes. The coordination entity alignment considers that the alignment between different entities is mutually influenced, and a global optimal alignment result is achieved by coordinating the matching condition between different objects.

The final stage of knowledge fusion is to resolve conflicts between different instances. The simplest way to identify conflicts is to find instances that differ for the same attributes and relationships. For the handling of conflicts, common strategies fall into the following three categories: conflict override, conflict avoidance, and conflict resolution. Conflict ignorance is not handled automatically, but is left to the user for resolution. Conflict avoidance does not resolve conflicts, but filters the data using rules or constraints. Conflict resolution focuses on resolving conflicts using features of the knowledge graph itself.

Referring to fig. 5, a schematic diagram of an example fusion of entity associations according to an embodiment of the present application is shown.

The event knowledge fusion adopts a fusion method based on entity association, a plurality of entities such as company names, industries, character entities and the like are often mentioned in time sequence and causal events, and the entities can be associated with entities in a knowledge graph or attributes of the entities, so that the fusion of the knowledge graph and the knowledge of the time sequence and causal events is realized, and the knowledge graph carries out reasoning and analysis by means of the logical affairs of the events. The section realizes named entity recognition by constructing an entity dictionary, and then fuses the knowledge map with the knowledge of time sequence and causal events through the association of the entities.

By named entity extraction, entities appearing in events are extracted, and next, the entities can be associated and matched with entities in a knowledge graph, so that a triple of < event entities, association relations and knowledge graph entities > is constructed to represent the association relations between the entities and the events. For example, a "Hua Yi sibling violent fall" event in a causal event and a stock entity "Hua Yi sibling" in a knowledge graph may constitute a < Hua Yi sibling violent fall, association, friendship sibling > triplet. Because the subject is to use a Neo4j graph database to store a knowledge graph, an event can be associated with an entity by simply establishing an edge between the event and the associated entity.

In the embodiment of the application, the subject selects a Neo4j graph database technology to store the knowledge graph. There are two main data types in Neo4j, nodes (nodes) and Edges (Edges), the nodes may correspond to entities in the knowledge graph, the Edges correspond to relationships between the entities in the knowledge graph, and attributes may be stored in both the nodes and the Edges.

In another feasible embodiment, in the field of supplier performance risk analysis, different knowledge maps are constructed by using different information sources, how to perform fusion expression on multiple knowledge maps is significant for establishing a uniform large-scale knowledge map. Because the information sources of different knowledge maps are different, the knowledge description systems of the knowledge maps are also different, semantically the same entities can have different expressions in different knowledge maps, and entities with the same name can represent different things. The multi-source knowledge graph fusion is not to simply combine knowledge graphs, but to discover equivalent examples, equivalent attributes or equivalent classes and the like among the knowledge graphs so as to realize entity alignment of the multi-source knowledge graphs.

According to the internal structure information (entities and relations) of the knowledge graph of the heterogeneous provider, a novel method of joint knowledge embedding is provided to achieve entity alignment of the multi-source knowledge graph in the provider performance risk analysis field, and alignment performance is improved based on an iterative training mode. The method comprises the steps of firstly, independently learning distributed expression of knowledge of different knowledge maps by using PTransE, jointly encoding entities and relations in the knowledge maps of heterogeneous suppliers into a uniform continuous low-dimensional semantic micro space, and then aligning the entities according to semantic distances among the entities in the united space. By researching the validity of the rich internal information of the knowledge graph on entity alignment, under the condition of giving a seed set for aligning the entities, the entities among different knowledge graphs are aligned only according to the internal structure of the knowledge graph.

According to the embodiment, the supplier fulfillment risk analysis method in the embodiment of the application constructs an ontology in a knowledge graph of a supplier according to the acquired supplier parameters; extracting the acquired supplier data based on a pre-trained entity extraction model to obtain supplier entity knowledge; extracting a text data set of a supplier by using a preset relation extraction frame to obtain a supplier entity relation; extracting the statement set of the supplier by using a pre-trained event extraction model to obtain supplier event knowledge; performing knowledge fusion based on the ontology, the supplier entity knowledge, the supplier entity relationship and the supplier event knowledge to obtain the supplier knowledge graph; and fusing the real-time important events of the suppliers based on the knowledge graph of the suppliers so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with the performing risk. By utilizing various technologies of entity extraction, relation extraction, event extraction, knowledge fusion and the like of natural language processing, internal and external data such as business license information, stockholder information, enterprise major personnel, external investment, judicial cases, patents, contract fulfillment and the like of a supplier unit are collected, a supplier fulfillment risk analysis library is established, and a knowledge map facing the supplier fulfillment risk audit is formed. And extracting the major time sequence and causal events of the supplier, and performing knowledge fusion with the knowledge map of the audit supplier to form clear evaluation and analysis data of the supplier, so as to realize the real-time holographic evaluation of the supplier and predict the fulfillment risk of the supplier.

It should be noted that the method of the embodiment of the present application may be executed by a single device, such as a computer or a server. The method of the embodiment can also be applied to a distributed scene and completed by the mutual cooperation of a plurality of devices. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiment, and the multiple devices interact with each other to complete the method.

It should be noted that the foregoing describes some embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments described above and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

Based on the same inventive concept, the application also provides a supplier performance risk analysis device corresponding to any embodiment method.

Referring to fig. 6, the supplier fulfillment risk analysis apparatus includes:

a construction module 61 configured to construct an ontology in the provider knowledge graph according to the obtained provider parameters;

an extraction module 62 configured to extract the obtained supplier data based on a pre-trained entity extraction model to obtain supplier entity knowledge;

the extraction module 63 is configured to extract the text data set of the provider by using a preset relationship extraction framework to obtain a provider entity relationship;

the extraction module 63 is further configured to extract the statement sets of the suppliers by using a pre-trained event extraction model, so as to obtain the supplier event knowledge;

a fusion module 64 configured to perform knowledge fusion based on the ontology, the supplier entity knowledge, the supplier entity relationship, and the supplier event knowledge to obtain the supplier knowledge graph;

the fusion module 64 is further configured to fuse real-time significant events of the suppliers based on the knowledge graph of the suppliers, so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with the risk of performing.

For convenience of description, the above devices are described as being divided into various modules by functions, which are described separately. Of course, the functionality of the various modules may be implemented in the same one or more pieces of software and/or hardware in the practice of the present application.

The apparatus of the foregoing embodiment is used to implement the corresponding supplier fulfillment risk analysis method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which are not described herein again.

Based on the same inventive concept, corresponding to any of the above-mentioned embodiments, the present application further provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the supplier fulfillment risk analysis method according to any of the above embodiments is implemented.

Fig. 7 is a schematic diagram illustrating a more specific hardware structure of an electronic device according to this embodiment, where the device may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein the processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 are communicatively coupled to each other within the device via a bus 1050.

The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits, and is configured to execute related programs to implement the technical solutions provided in the embodiments of the present disclosure.

The Memory 1020 may be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, or the like. The memory 1020 may store an operating system and other application programs, and when the technical solution provided by the embodiments of the present specification is implemented by software or firmware, the relevant program codes are stored in the memory 1020 and called to be executed by the processor 1010.

The input/output interface 1030 is used for connecting an input/output module to input and output information. The i/o module may be configured as a component in a device (not shown) or may be external to the device to provide a corresponding function. The input devices may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices may include a display, a speaker, a vibrator, an indicator light, etc.

The communication interface 1040 is used for connecting a communication module (not shown in the drawings) to implement communication interaction between the present device and other devices. The communication module can realize communication in a wired mode (for example, USB, network cable, etc.), and can also realize communication in a wireless mode (for example, mobile network, WIFI, bluetooth, etc.).

The bus 1050 includes a path to transfer information between various components of the device, such as the processor 1010, memory 1020, input/output interface 1030, and communication interface 1040.

It should be noted that although the above-mentioned device only shows the processor 1010, the memory 1020, the input/output interface 1030, the communication interface 1040 and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, those skilled in the art will appreciate that the above-described apparatus may also include only the components necessary to implement the embodiments of the present disclosure, and need not include all of the components shown in the figures.

The electronic device of the foregoing embodiment is used to implement the corresponding supplier fulfillment risk analysis method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which are not described herein again.

Based on the same inventive concept, corresponding to any of the above-described embodiment methods, the present application further provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the supplier fulfillment risk analysis method according to any of the above embodiments.

Computer-readable media of the present embodiments, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

The storage medium of the above embodiment stores computer instructions for causing the computer to execute the provider fulfillment risk analysis method according to any of the above embodiments, and has the beneficial effects of corresponding method embodiments, which are not described herein again.

Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, is limited to these examples; within the context of the present application, technical features in the above embodiments or in different embodiments may also be combined, steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application described above, which are not provided in detail for the sake of brevity.

In addition, well-known power/ground connections to Integrated Circuit (IC) chips and other components may or may not be shown in the provided figures for simplicity of illustration and discussion, and so as not to obscure the embodiments of the application. Furthermore, devices may be shown in block diagram form in order to avoid obscuring embodiments of the application, and this also takes into account the fact that specifics with respect to implementation of such block diagram devices are highly dependent upon the platform within which the embodiments of the application are to be implemented (i.e., specifics should be well within purview of one skilled in the art). Where specific details (e.g., circuits) are set forth in order to describe example embodiments of the application, it should be apparent to one skilled in the art that the embodiments of the application can be practiced without, or with variation of, these specific details. Accordingly, the description is to be regarded as illustrative instead of restrictive.

While the present application has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art in light of the foregoing description. For example, other memory architectures, such as Dynamic RAM (DRAM), may use the discussed embodiments.

The present embodiments are intended to embrace all such alternatives, modifications and variances which fall within the broad scope of the appended claims. Therefore, any omissions, modifications, substitutions, improvements, and the like that may be made without departing from the spirit and principles of the embodiments of the present application are intended to be included within the scope of the present application.

Claims

1. A supplier fulfillment risk analysis method, comprising:

extracting the acquired supplier data based on a pre-trained entity extraction model to obtain supplier entity knowledge;

and fusing real-time important events of the suppliers based on the knowledge graph of the suppliers so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with the performing risk.

2. The method of claim 1, wherein constructing an ontology in a provider knowledge graph based on the obtained provider parameters comprises:

defining the application field and knowledge range of the ontology;

defining attributes and attribute values of entities in the ontology;

3. The method of claim 1, wherein the training process of the pre-trained entity extraction model comprises:

acquiring an original supplier data set;

4. The method of claim 1, wherein extracting the text data set of the supplier by using the preset relationship extraction framework to obtain the supplier entity relationship comprises:

and in response to the scale of the second entity relationship instance set reaching a preset second threshold, taking the second entity relationship instance set as the provider entity relationship.

5. The method of claim 1, wherein the extracting the statement set of the supplier using the pre-trained event extraction model to obtain the supplier event knowledge comprises:

in the pre-trained event extraction model, expanding each sentence in the sentence set of the supplier to a fixed preset length to obtain a sentence set with a preset length; coding each word in each sentence in the preset length sentence set to obtain a word vector set with fixed dimensions; and capturing context information of statements in the statement set of the supplier based on the word vector set with fixed dimensions, and performing global optimization on the context information by using a fixed rate coefficient layer to obtain the supplier event knowledge.

6. The method of claim 1, wherein the knowledge fusion comprises: entity knowledge fusion and event knowledge fusion;

the knowledge fusion is performed based on the ontology, the supplier entity knowledge, the supplier entity relationship and the supplier event knowledge to obtain the supplier knowledge graph, which includes:

7. The method of claim 6, further comprising:

8. A supplier fulfillment risk analysis apparatus, comprising:

the extraction module is configured to extract a text data set of a supplier by using a preset relation extraction frame to obtain a supplier entity relation;

and the fusion module is also configured to fuse real-time important events of the suppliers based on the knowledge graph of the suppliers so as to analyze the risk that the suppliers cannot perform and obtain the suppliers with the performing risk.

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 7 when executing the program.

10. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1 to 7.