Industrial chain knowledge graph expansion method, electronic device and storage medium

Abstract

The present invention provides an industrial chain knowledge graph expansion method, electronic device and storage medium, the method comprising: extracting a number of first product keywords according to a number of product description texts of a given product in a corresponding enterprise object, then obtaining a number of target texts corresponding to the screened second product keywords from a preset text library, obtaining a target category label corresponding to the given product according to the number of target texts, and obtaining a target industrial chain corresponding to the enterprise object according to the target category label mapping, obtaining the number of second product keywords whose keyword similarity with any third product keyword is greater than a preset similarity threshold, and when the ratio of the number to the total number of second product keywords is greater than a preset ratio threshold, the enterprise object, the given product and the target chain point in the target industrial chain are fused and connected; the present invention can realize accurate and reliable expansion of the industrial chain knowledge graph.

Description

Technical Field

[0001] The present invention relates to the field of text processing technology, and in particular to an industrial chain knowledge graph expansion method, electronic equipment and storage medium. Background Art

[0002] The industrial chain knowledge graph includes the triple relationship between enterprises and industrial chain nodes, enterprises and products, and nodes and nodes. By extracting the entity and attribute relationship in the enterprise industrial chain knowledge graph, the association relationship between several enterprise objects and products can be obtained. Enterprise industrial chain information has a vital reference value in many aspects such as enterprise risk assessment, risk transmission and industry correlation analysis. In some existing public documents of enterprises, the industrial chain relationship of some enterprises associated with them is often revealed. For example, in the public documents released by the enterprise, users can view the material source and sales destination of the products sold by the enterprise, so as to determine some enterprises and products associated with the enterprise. However, based on these public documents, only a one-sided understanding of the industrial chain relationship can be made, and due to the lack of data in the industrial chain knowledge graph, it is impossible to obtain comprehensive industrial chain information, and the reliability of enterprise risk assessment and industry correlation analysis cannot be guaranteed. Summary of the invention

[0003] In view of the above technical problems, the technical solution adopted by the present invention is:

[0004] A method for expanding an industrial chain knowledge graph, the method comprising the following steps:

[0005] For a given product of an enterprise object, a number of first product keywords are extracted through a keyword extraction model according to a number of product description texts in the enterprise object corresponding to the given product.

[0006] According to the TF-IDF value corresponding to each first product keyword and a preset TF-IDF threshold, a plurality of second product keywords are screened out from the plurality of first product keywords, and a plurality of target texts corresponding to the plurality of second product keywords are obtained from a preset text library.

[0007] A number of third product keywords are extracted from a number of target texts, and feature vectors are generated and input into the target neural network prediction model to obtain the target category label corresponding to the given product, so as to obtain the target industrial chain corresponding to the enterprise object according to a preset industrial chain mapping table; the industrial chain mapping table includes a number of groups of one-to-one corresponding preset category labels and preset industrial chains.

[0008] The keyword similarity between any second product keyword and each third product keyword is calculated, and when there is a keyword similarity greater than a preset similarity threshold, it is determined that the second product keyword itself and the corresponding third product keyword have a matching relationship.

[0009] When the ratio of the number of second product keywords with matching relationships to the total number of second product keywords is greater than a preset ratio threshold, the enterprise object, the given product and the target chain point are fused and linked according to the target chain point corresponding to the given product in the target industrial chain, so as to achieve the expansion of the industrial chain knowledge graph.

[0010] The present invention has at least the following beneficial effects:

[0011] The present invention provides an industrial chain knowledge graph expansion method. First, according to a number of product description texts of a given product in a corresponding enterprise object, a number of first product keywords are extracted. Then, according to the importance of each first product keyword, a number of second product keywords are screened out. The target text corresponding to the second product keyword is obtained from a preset text library. A number of third product keywords are extracted from the target texts to improve the comprehensiveness of the vocabulary related to the given product. According to the number of third product keywords, a target category label corresponding to the given product is obtained. According to the target category label mapping, a target industrial chain corresponding to the enterprise object is obtained, that is, the industrial chain corresponding to the industry to which the enterprise object belongs is obtained. The number of second product keywords whose keyword similarity with any third product keyword is greater than a preset similarity threshold is obtained. When the ratio of the number to the total number of second product keywords is greater than a preset ratio threshold, it indicates that the same number or similar number of the second product keywords and the third product keywords are large, and the obtained target industrial chain can be reasonably determined as the industrial chain corresponding to the given product. At this time, the enterprise object, the given product and the target chain point in the target industrial chain are fused and connected, so that the accurate and reliable expansion of the industrial chain knowledge graph can be achieved, which is beneficial to the enterprise object's understanding of all-domain products and other enterprise objects, so as to facilitate subsequent enterprise risk assessment and industry correlation analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 A flowchart of an industrial chain knowledge graph expansion method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0015] This embodiment provides a method for expanding an industrial chain knowledge graph, and the method includes the following steps:

[0016] S100, for a given product of an enterprise object, extracting a plurality of first product keywords through a keyword extraction model according to a plurality of product description texts in the enterprise object corresponding to the given product.

[0017] Specifically, the product description text may be a patent text or a scientific research project text corresponding to a given product in an enterprise object.

[0018] Preferably, the keyword extraction model is a TextRank model, which can accurately extract the first product keyword based on the semantic information of the product description text; those skilled in the art are aware of the specific implementation process of extracting keywords using the TextRank model in the prior art, which will not be repeated here.

[0019] Furthermore, the step of extracting a plurality of first product keywords through the keyword extraction model further includes the following steps:

[0020] S101, when the number of first product keywords extracted from a plurality of product description texts is greater than a preset number threshold, a preset grammar filter is used to filter the plurality of first product keywords to obtain a new keyword vertex set.

[0021] S102: Determine a plurality of first product keywords in the new keyword vertex set as a plurality of first product keywords finally extracted.

[0022] As mentioned above, when the number of extracted first product keywords is large, the first product keywords with low importance can be screened out through the processing of the preset grammar filter, thereby achieving the optimization of the first product keyword set. By extracting a number of first product keywords, it is beneficial to subsequently obtain a number of target texts related to a given product.

[0023] S200, based on the TF-IDF value corresponding to each first product keyword and the preset TF-IDF threshold, select a number of second product keywords from a number of first product keywords, and obtain a number of target texts corresponding to the number of second product keywords from a preset text library; it can be understood as: the first product keyword whose corresponding TF-IDF value is greater than the preset TF-IDF threshold is determined as the second product keyword.

[0024] In a specific embodiment, in step S200, the target text is obtained by the following steps:

[0025] S201, obtaining a plurality of first texts to be processed corresponding to each second product keyword from a preset text library, and determining a plurality of target sentences from each first text to be processed; the target sentence refers to a sentence including at least one second product keyword obtained from a target paragraph in any first text to be processed.

[0026] In a specific embodiment, in step S201, the target paragraph in the first to-be-processed text is determined by the following steps:

[0027] S2011, obtaining from the preset text library a number of first texts to be processed corresponding to each second product keyword and a text category identifier of each first text to be processed; this can be understood as: obtaining from the preset text library a text containing at least one second product keyword as the first text to be processed.

[0028] Specifically, the text category identification includes but is not limited to patent text and paper text.

[0029] S2012, for any first text to be processed, determine the target paragraph from the first text to be processed according to the correspondence between the preset text category identifier and the target paragraph in the first text to be processed; for example, when the text category identifier is a patent text, considering the repetitiveness of words, the specific implementation method part is taken as the target paragraph.

[0030] In the above, by obtaining the first text to be processed based on several second product keywords, the obtained related text is made more comprehensive, and then the target sentence containing the second product keywords is further determined from the target paragraph of the first text to be processed, so that the number of each second product keyword in the obtained target sentence can provide a basis for calculating the text priority of the first text to be processed.

[0031] S202: For any first text to be processed, according to the quantity information of each second product keyword in each target sentence in the first text to be processed, obtain the first text priority corresponding to the first text to be processed.

[0032] Specifically, the first text priority corresponding to the first to-be-processed text is obtained through the following steps:

[0033] S2021. For any target sentence in the first text to be processed, divide the several second product keywords in the target sentence into several target phrases; the target phrase refers to any phrase in which the number of second product keywords screened out among the currently remaining several second product keywords is the largest and each screened out second product keyword has only one; the currently remaining several second product keywords refer to the remaining several second product keywords obtained after deleting all the screened out second product keywords from the several second product keywords in the target sentence.

[0034] For a better understanding, an example is given below: when the second product keywords in the target sentence include A1, A2, A3, A2, A1, A1, A1, the first screening obtains the target phrase (A1, A2, A3), the remaining second product keywords currently include A2, A1, A1, A1, the second screening obtains the target phrase (A1, A2), the remaining second product keywords currently include A1, A1, the third screening obtains the target phrase (A1), the remaining second product keywords currently are A1, so the last screening obtains the target phrase (A1).

[0035] S2022, for any target phrase corresponding to the target sentence, calculate the first phrase priority corresponding to the target phrase based on the total number of second product keywords in the target phrase and the target number corresponding to the target phrase; the target number corresponding to the target phrase refers to the number of phrases obtained from several target phrases corresponding to the target sentence that are the same as any target phrase.

[0036] Further, the first phrase priority corresponding to the target phrase is the product of the total number of second product keywords in the target phrase and the target number corresponding to the target phrase.

[0037] For example, when the target phrase is (A1, A2, A3), the total number of second product keywords in the target phrase is 3, the target number corresponding to the target phrase is 1, and the priority of the first phrase corresponding to the target phrase is 3; when the target phrase is (A1), the total number of second product keywords in the target phrase is 1, the target number corresponding to the target phrase is 2, and the priority of the first phrase corresponding to the target phrase is 2.

[0038] S2023, based on the same target phrase information in the several target phrases corresponding to the target sentence, deduplicate the several first phrase priorities to obtain several second phrase priorities; it can be understood as: deduplicate the target phrases, and also remove the first phrase priorities corresponding to the removed target phrases, because when calculating the first phrase priority of the target phrase, all target phrases that are the same as the target phrase have been taken into account.

[0039] S2024: Determine the sum of the priorities of the plurality of second phrases as the sentence priority corresponding to the target sentence.

[0040] S2025, adding sentence priorities corresponding to a plurality of target sentences in the first text to be processed, to obtain a first text priority corresponding to the first text to be processed.

[0041] As described above, when calculating the first text priority of the first text to be processed, the dimension of the second product keywords contained in the first text to be processed is considered. By introducing the combination of the second product keywords and the number of each combination, the importance of the first text to be processed can be more accurately reflected.

[0042] S203, obtaining a number of second texts to be processed corresponding to the first text to be processed, and obtaining the second text priority corresponding to the first text to be processed based on a number of second product keywords corresponding to each second text to be processed and the length of time between the publication time of the first text to be processed and the publication time of each second text to be processed.

[0043] Specifically, the second text to be processed refers to any first text to be processed whose publication time is earlier than that of the first text to be processed.

[0044] Specifically, the second text priority corresponding to the first to-be-processed text is obtained through the following steps:

[0045] S2031, normalize each time length according to the time length between the first to-be-processed text publication time and each second to-be-processed text publication time, and obtain a time proportion parameter corresponding to each second to-be-processed text.

[0046] S2032, for any second text to be processed, according to the second product keywords corresponding to the second text to be processed, obtain the number of second product keywords repeated between the second text to be processed and the first text to be processed. For example, the first text to be processed includes the second product keywords A1, A2, A3, and the second text to be processed includes the second product keywords A1 and A2, then the total number of repeated second product keywords is 2.

[0047] S2033: Determine the product of the time proportion parameter corresponding to the second text to be processed and the total number of second product keywords repeated between the second text to be processed and the first text to be processed as the text weight of the second text to be processed.

[0048] S2034, calculating the ratio of the sum of the text weights of each second text to be processed to the number of second texts to be processed, and using the calculation result as the second text priority corresponding to the first text to be processed.

[0049] As described above, when calculating the second text priority of the first text to be processed, the dimension of several second texts to be processed whose publication time is before the first text to be processed is taken into consideration, and the time dimension is also introduced into the number of repeated words as the weight of the number of repeated words, which can accurately reflect the importance of each second text to be processed to the first text to be processed, thereby improving the reliability of the second text priority corresponding to the first text to be processed.

[0050] S204: When the sum of the first text priority and the second text priority corresponding to the first text to be processed is greater than a preset priority threshold, the first text to be processed is used as the target text.

[0051] As described above, the priority of the text is considered through multiple dimensions, which makes the determination of the target text with high relevance to the second product keyword more reasonable and reliable, and the several target texts with different disclosure times that are finally determined are helpful to reflect the development progress of the product to a certain extent, making the extraction of the third product keyword more comprehensive.

[0052] S300, extracting several third product keywords from several target texts, and generating feature vectors to be input into the target neural network prediction model, obtaining the target category label corresponding to the given product, and obtaining the target industrial chain corresponding to the enterprise object according to the preset industrial chain mapping table; the industrial chain mapping table includes several groups of one-to-one corresponding preset category labels and preset industrial chains.

[0053] Specifically, the preset industrial chain includes several industrial chain points from upstream to downstream, such as production raw materials, auxiliary materials, production equipment, auxiliary equipment, processing technology, technical services, product business, and sales channels.

[0054] In a specific embodiment, the target neural network prediction model is obtained by the following steps:

[0055] S10, configuring the neurons of the output layer of the preset neural network prediction model according to the number of preset category labels; it can be understood that: the number of neurons in the output layer is equal to the number of preset category labels.

[0056] Specifically, the preset category label refers to a label used to characterize the industry to which the enterprise belongs based on the production scope of the enterprise.

[0057] S20, obtaining a number of historical products under each preset category label and a number of historical texts corresponding to each historical product; the method for obtaining the historical texts is consistent with the method for obtaining the target text.

[0058] S30, for any historical product under any preset category label, extract several historical product keywords from several historical texts corresponding to the historical product, and input the several historical product keywords into the preset neural network prediction model for feature extraction to obtain a text feature vector; it can be understood as: merging several historical product keywords into one text for feature extraction.

[0059] S40, using a number of text feature vectors to train a preset neural network prediction model, and obtaining a target category label corresponding to each historical product from an output layer.

[0060] S50, iteratively adjusting the parameters of the preset neural network prediction model according to the target category label corresponding to each historical product and the preset category label corresponding to each historical product, so as to construct a target neural network prediction model.

[0061] As described above, through historical products of different categories and several historical texts corresponding to each historical product, a large number of text features are extracted to train the preset neural network prediction model, so that the trained target neural network prediction model can accurately and reliably predict the category to which the product belongs.

[0062] S400, calculating the keyword similarity between any second product keyword and each third product keyword, and when there is a keyword similarity greater than a preset similarity threshold, determining that the second product keyword itself and the corresponding third product keyword have a matching relationship.

[0063] Specifically, the keyword similarity between any second product keyword and any third product keyword meets the following conditions:

[0064] ζ=(α·β) / (||α||×||β||), where ζ is the keyword similarity between any second product keyword and any third product keyword, α is the word vector corresponding to any second product keyword, and β is the word vector corresponding to any third product keyword.

[0065] As mentioned above, since the second product keyword and the third product keyword are not exactly the same, they may also represent the same object or have the same semantics, so the keyword similarity between the two is calculated. When the keyword similarity is large, it is considered that there is a matching relationship, which makes the judgment of the overlapping amount of the second product keyword and the third product keyword more accurate, thereby providing a judgment basis for determining whether the target industrial chain is the industrial chain corresponding to a given product.

[0066] S500, when the ratio of the number of second product keywords with matching relationships to the total number of second product keywords is greater than a preset ratio threshold, the enterprise object, the given product and the target chain point are fused and linked according to the target chain point corresponding to the given product in the target industrial chain, so as to expand the knowledge graph of the industrial chain.

[0067] As mentioned above, when the ratio of the number of second product keywords with matching relationships to the total number of second product keywords is large, it indicates that the number of identical or similar second product keywords and third product keywords is relatively large, and the acquired target industrial chain can be reasonably determined as the industrial chain corresponding to the given product. At this time, the enterprise object, the given product and the target chain point in the target industrial chain are integrated and connected, which can realize the accurate and reliable expansion of the industrial chain knowledge graph, and then help the enterprise object to understand the full range of products and other enterprise objects, so as to facilitate the subsequent enterprise risk assessment and industry correlation analysis.

[0068] In a specific embodiment, the target chain point corresponding to a given product in the target industrial chain is determined by the following steps:

[0069] S001, obtain the name of each industrial chain point in the target industrial chain.

[0070] S002, use the NLP model to perform semantic analysis on several product description texts, determine the semantic text similarity between any product description text and each industrial chain link name, and determine the maximum semantic text similarity corresponding to any product description text. For example, when there are eight industrial chain link names, one product description text corresponds to eight semantic text similarities, and the maximum value among them is used as the maximum semantic text similarity corresponding to the product description text; the NLP model obtains semantic text similarity as a prior art, which will not be repeated here.

[0071] S003, determining the industrial chain point with the largest number of corresponding maximum semantic text similarities as the target link point corresponding to the given product in the target industrial chain.

[0072] In the above, by analyzing the semantic text similarities between several product description texts corresponding to the same product and each industrial chain point, and according to the number of maximum semantic text similarities corresponding to the industrial chain points, the target chain point corresponding to a given product is determined, so that the obtained target chain point is more accurate and reliable.

[0073] The present invention also provides a method for identifying upstream and downstream relationships of enterprises. In one embodiment, firstly, upstream and downstream identification requests of two enterprise objects are received. When both enterprise objects exist in the same target industrial chain in the industrial chain knowledge graph, the upstream and downstream information of the two enterprise objects are directly obtained. In another embodiment, when at least one of the two enterprise objects does not exist in the industrial chain knowledge graph, the method for identifying upstream and downstream relationships of enterprises includes the following steps:

[0074] P100, based on the enterprise information corresponding to several given enterprise objects and the preset chain point prediction model, obtain the industrial chain chain point corresponding to each enterprise object; the enterprise information corresponding to the enterprise object includes the enterprise object name and the product introduction information of the enterprise object; it can be understood that: the product introduction information of the enterprise object is the names of several business products of the enterprise object.

[0075] In a specific embodiment, the method constructs the chain point prediction model through the following steps:

[0076] P001, configure the output layer of a given neural network prediction model according to a number of industrial chain points in a given industrial chain; it can be understood that the number of neurons in the output layer is equal to the number of industrial chain points.

[0077] P002, input the pre-acquired enterprise information sample set into the given neural network prediction model for text feature extraction, and obtain a number of text feature vectors; the enterprise information sample set includes the name of the enterprise object sample and the product introduction information of the enterprise object sample; it can be understood as: the name of the enterprise object sample and the product introduction information of the enterprise object sample corresponding to the same enterprise information sample are merged into one text for feature extraction.

[0078] P003, uses several text feature vectors to train the given neural network prediction model, and obtains the predicted industrial chain point corresponding to each enterprise object sample from the output layer.

[0079] P004, according to the predicted industrial chain points and actual industrial chain points corresponding to each enterprise object sample, iteratively adjust the parameters of the given neural network prediction model to construct a chain point prediction model.

[0080] As mentioned above, since the enterprise object name and the product introduction information of the enterprise object can well reflect the production scope of the enterprise object, the chain point prediction model is trained by extracting text features of the two, so that the trained chain point prediction model can more accurately predict the production situation of the enterprise object, and then determine the chain point of the industrial chain to which the enterprise object belongs.

[0081] P200, based on a number of preset category labels and according to the product introduction information of each enterprise object, a number of enterprise objects are clustered to determine the category label of each enterprise object.

[0082] In a specific embodiment, the P200 step includes the following steps:

[0083] P201, according to the number of preset category labels, construct the same number of clusters to be clustered, and use the word vector of each preset category label as the center of each cluster to be clustered.

[0084] P202, for any enterprise object, extract several enterprise object keywords from the product introduction information of the enterprise object and form an enterprise keyword group; it can be understood as: extracting several enterprise object keywords through a keyword extraction model.

[0085] S203, calculating the distance between the phrase vector of the enterprise keyword group and the center of each cluster to be clustered, and determining the target cluster center; the target cluster center refers to the cluster center to be clustered that is closest to the phrase vector of the enterprise keyword group. For example, the phrase vectors of several enterprise keyword groups are clustered using a k-means clustering model.

[0086] P204, determining the preset category label corresponding to the target cluster center as the category label of the enterprise object.

[0087] As mentioned above, since the product introduction information can reflect the product range produced by the enterprise, several enterprise object keywords are extracted from it and formed into enterprise keyword groups. By clustering the enterprise keyword groups, enterprise objects of the same category can be more accurately clustered into the same cluster, thereby obtaining reliable category labels for the enterprise objects.

[0088] P300, for the target product of any enterprise object, according to several product description texts in the enterprise object corresponding to the target product, several target product keywords are extracted through the keyword extraction model to obtain the associated product tags and associated technical tags of the target product.

[0089] In a specific embodiment, the P300 step includes the following steps:

[0090] P301, calculate the TF-IDF value corresponding to each target product keyword, and determine the target product keyword whose corresponding TF-IDF value is greater than a preset TF-IDF threshold as a target product associated word.

[0091] P302, use the trained natural language processing model to obtain product entity words and technical entity words from the target product related words, and use the product entity words and technical entity words as the associated product labels and associated technology labels of the target product respectively.

[0092] Specifically, the natural language processing model is trained based on a preset corpus, wherein the preset corpus includes a number of product entity words and a number of technical entity words; it can be understood as: directly extracted product entity words or product entity words with the same semantics as any preset product entity words are used as associated product labels, and directly extracted technical entity words or technical entity words with the same semantics as any preset technical entity words are used as associated technical labels.

[0093] As mentioned above, by obtaining the associated product tags and associated technology tags of the target product, it is possible to reflect the products or technologies associated with the target product in any enterprise object. Since the upstream and downstream enterprises have overlapping products or technologies, the associated product tags and associated technology tags are used as two associated parameters of the target product, which is conducive to accurately obtaining the association similarity between the two enterprise objects.

[0094] P400, calculates the label similarity corresponding to the category label, associated product label and associated technology label between every two enterprise objects, and weights and calculates the label similarity according to the preset label weights corresponding to the category label, associated product label and associated technology label to obtain the associated similarity between every two enterprise objects.

[0095] In a specific embodiment, step P400 further includes the following steps:

[0096] P401, for the same cluster center to be clustered, obtain several enterprise keyword groups corresponding to the cluster center to be clustered; it can be understood that: the several enterprise keyword groups corresponding to the cluster center to be clustered refer to several enterprise keyword groups whose corresponding phrase vectors are closest to the cluster center to be clustered.

[0097] P402, for any enterprise keyword group corresponding to the center of the cluster to be clustered, according to the distance between the phrase vector of the enterprise keyword group and the center of the cluster to be clustered, a target distance interval corresponding to the enterprise keyword group is mapped from a plurality of preset distance intervals.

[0098] P403, according to the comparison table of preset distance intervals and preset weight coefficients, the preset weight coefficient corresponding to the target distance interval is determined as the preset weight coefficient corresponding to the enterprise keyword group.

[0099] Specifically, the comparison table of preset distance intervals and preset weight coefficients refers to a data table recording a number of preset distance intervals and a number of preset weight coefficients, and the preset distance intervals and the preset weight coefficients correspond one to one.

[0100] P404, when calculating the association similarity between two enterprise objects, the preset tag weight corresponding to the category tag is obtained according to the preset weight coefficients corresponding to the enterprise keyword groups of the two enterprise objects themselves and the given tag weight corresponding to the category tag.

[0101] As mentioned above, since there will inevitably be certain errors when obtaining the category labels of enterprise objects through clustering, the clustering results are further analyzed, and the accuracy of the category labels is measured according to the distance between the phrase vector of each enterprise keyword group and the center of the corresponding cluster. The closer the distance, the more accurate the clustering result. Therefore, the preset weight coefficient obtained according to the distance is assigned to the given label weight corresponding to the category label, so that the weights assigned to the category labels of the two enterprise objects are more reasonable, thereby improving the accuracy of the similarity calculation.

[0102] P500, when the association similarity between two enterprise objects is greater than the preset similarity threshold, the upstream and downstream relationship between the two enterprise objects is identified according to the industrial chain points corresponding to the two enterprise objects respectively; it can be understood that: when the association similarity between two enterprise objects is greater than the preset similarity threshold, it is determined that the two enterprise objects have an association relationship.

[0103] As mentioned above, after knowing the industrial chain points corresponding to the two enterprise objects, since the two enterprise objects may not belong to the same industry, it is impossible to determine whether the two enterprise objects have an upstream and downstream relationship. By introducing the category label, associated product label and associated technology label of each enterprise object, the association similarity between the two enterprise objects is calculated, so that the calculated association similarity is more accurate and reliable, and then it is known that the two enterprise objects have an association relationship, and the upstream and downstream relationship between the two enterprise objects is determined through the corresponding industrial chain points, which is conducive to subsequent enterprise risk assessment and industry correlation analysis.

[0104] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which can be set in an electronic device to store at least one instruction or at least one program related to implementing a method in a method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the industrial chain knowledge graph expansion method provided in the above embodiment.

[0105] An embodiment of the present invention further provides an electronic device, comprising a processor and the aforementioned non-transitory computer-readable storage medium.

[0106] An embodiment of the present invention further provides a computer program product, which includes program code. When the program product is run on an electronic device, the program code is used to enable the electronic device to execute the steps of the method according to various exemplary embodiments of the present invention described above in this specification.

[0107] Although some specific embodiments of the present invention have been described in detail by way of example, it will be appreciated by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It will also be appreciated by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A method for expanding an industrial chain knowledge graph, characterized in that: The method comprises the following steps: For a given product of an enterprise object, extracting a number of first product keywords through a keyword extraction model according to a number of product description texts in the enterprise object corresponding to the given product; According to the TF-IDF value corresponding to each first product keyword and a preset TF-IDF threshold, a plurality of second product keywords are selected from the plurality of first product keywords, and a plurality of target texts corresponding to the plurality of second product keywords are obtained from a preset text library; Extracting a number of third product keywords from a number of target texts, and generating feature vectors to be input into a target neural network prediction model, obtaining a target category label corresponding to a given product, and obtaining a target industrial chain corresponding to the enterprise object according to a preset industrial chain mapping table; the industrial chain mapping table includes a number of sets of one-to-one corresponding preset category labels and preset industrial chains; Calculate the keyword similarity between any second product keyword and each third product keyword, and when there is a keyword similarity greater than a preset similarity threshold, determine that the second product keyword itself and the corresponding third product keyword have a matching relationship; When the ratio of the number of second product keywords with matching relationships to the total number of second product keywords is greater than a preset ratio threshold, the enterprise object, the given product and the target chain point are fused and connected according to the target chain point corresponding to the given product in the target industrial chain, so as to achieve the expansion of the industrial chain knowledge graph; The target text is obtained through the following steps: Acquire a plurality of first to-be-processed texts corresponding to each second product keyword from a preset text library, and determine a plurality of target sentences from each first to-be-processed text; the target sentence refers to a sentence obtained from a target paragraph in any first to-be-processed text and including at least one second product keyword; For any first text to be processed, according to the quantity information of each second product keyword in each target sentence in the first text to be processed, obtaining the first text priority corresponding to the first text to be processed; Obtaining a number of second texts to be processed corresponding to the first text to be processed, and obtaining the second text priority corresponding to the first text to be processed according to a number of second product keywords corresponding to each second text to be processed and the time between the first text to be processed and the time between the first text to be processed and the time between the first text to be processed and the time between the second text to be processed; the second text to be processed refers to any first text to be processed whose time of publication is earlier than that of the first text to be processed; When the sum of the first text priority and the second text priority corresponding to the first text to be processed is greater than a preset priority threshold, the first text to be processed is used as the target text.

2. The method for expanding the industrial chain knowledge graph according to claim 1, characterized in that: The step of extracting a plurality of first product keywords through the keyword extraction model further includes the following steps: When the number of first product keywords extracted from the product description texts is greater than a preset number threshold, a preset grammar filter is used to filter the first product keywords to obtain a new keyword vertex set; The first product keywords in the new keyword vertex set are determined as the first product keywords finally extracted.

3. The method for expanding the industrial chain knowledge graph according to claim 1, characterized in that: The first text priority corresponding to the first to-be-processed text is obtained by following the steps below: For any target sentence in the first to-be-processed text, the second product keywords in the target sentence are divided into a plurality of target phrases; the target phrase refers to any phrase in which the number of second product keywords screened out is the largest among the currently remaining second product keywords and each screened out second product keyword has only one; the currently remaining second product keywords refer to the remaining second product keywords obtained after deleting all the screened out second product keywords from the second product keywords in the target sentence; For any target phrase corresponding to the target sentence, the first phrase priority corresponding to the target phrase is calculated according to the total number of second product keywords in the target phrase and the target number corresponding to the target phrase; the target number corresponding to the target phrase refers to the number of phrases obtained from the target phrases corresponding to the target sentence that are the same as any target phrase; According to the same target phrase information in the target phrases corresponding to the target sentence, a plurality of first phrase priorities are deduplicated to obtain a plurality of second phrase priorities; Determining the sum of the priorities of the plurality of second phrases as the sentence priority corresponding to the target sentence; The sentence priorities corresponding to several target sentences in the first text to be processed are added together to obtain a first text priority corresponding to the first text to be processed.

4. The method for expanding the industrial chain knowledge graph according to claim 1, characterized in that: The second text priority corresponding to the first to-be-processed text is obtained by following the steps below: According to the time between the publication time of the first to-be-processed text and the publication time of each second to-be-processed text, each time is normalized to obtain a time proportion parameter corresponding to each second to-be-processed text; For any second text to be processed, according to a number of second product keywords corresponding to the second text to be processed, the number of second product keywords repeated between the second text to be processed and the first text to be processed is obtained; Determine the product of the time proportion parameter corresponding to the second text to be processed and the total number of second product keywords repeated between the second text to be processed and the first text to be processed as the text weight of the second text to be processed; The ratio of the sum of the text weights of each second text to be processed to the number of the second texts to be processed is calculated, and the calculation result is used as the second text priority corresponding to the first text to be processed.

5. The method for expanding the knowledge graph of the industrial chain according to claim 1, characterized in that: Obtain the target neural network prediction model through the following steps: According to the number of preset category labels, the neurons of the output layer of the preset neural network prediction model are configured; Acquire a number of historical products under each preset category label and a number of historical texts corresponding to each historical product; the method for acquiring the historical texts is consistent with the method for acquiring the target text; For any historical product under any preset category label, extract a number of historical product keywords from a number of historical texts corresponding to the historical product, and input the number of historical product keywords into a preset neural network prediction model for feature extraction to obtain a text feature vector; Use several text feature vectors to train the preset neural network prediction model, and obtain the target category label corresponding to each historical product from the output layer; According to the target category label corresponding to each historical product and the preset category label corresponding to each historical product, the preset neural network prediction model is iteratively adjusted to construct a target neural network prediction model.

6. The method for expanding the industrial chain knowledge graph according to claim 1, characterized in that: The keyword similarity between any second product keyword and any third product keyword meets the following conditions: ζ=(α·β) / (||α||×||β||), where ζ is the keyword similarity between any second product keyword and any third product keyword, α is the word vector corresponding to any second product keyword, and β is the word vector corresponding to any third product keyword.

7. The method for expanding the knowledge graph of the industrial chain according to claim 1, characterized in that: Determine the target chain point corresponding to a given product in the target industrial chain through the following steps: Obtain the names of each industrial chain point in the target industrial chain; Use the NLP model to perform semantic analysis on several product description texts, determine the semantic text similarity between any product description text and the name of each industrial chain link, and determine the maximum semantic text similarity corresponding to any product description text; The industrial chain point with the largest number of corresponding maximum semantic text similarities is determined as the target link point corresponding to the given product in the target industrial chain.

8. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the storage medium, characterized in that: The at least one instruction or the at least one program is loaded and executed by the processor to implement the industrial chain knowledge graph expansion method as described in any one of claims 1-7.

9. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 8.

Images