A method, device and medium for automatically generating document labels and an electronic device

Abstract

The application discloses a document label automatic generation method, device and medium and an electronic device. The document label automatic generation method first extracts keywords based on an input document, then performs domain mapping on the keywords, determines a focus domain through analysis of the relevance of the domain and the document, and then generates a label based on the focus domain. The application can shield the interference of non-focus domain keywords, accurately determine the content center of the document, and thus accurately determine the corresponding label. The final actual technical effect of the application mainly depends on the construction of the input domain label tree. As long as the domain label tree is configured accurately enough, the determined label is also accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to internet data analysis, in particular to the automation of document tag generation. BACKGROUND

[0002] Technical documents exchanged through the internet need to be tagged or keyworded for easy retrieval. Technical documents can be text-based or multimedia such as pictures, audio or video. If each document needs to be manually tagged or keyworded, the workload is huge, and manual tagging is prone to errors, inaccuracy, subjectivity and lack of standardization. Therefore, a technical document retrieval center needs an automated document tag generation tool. SUMMARY

[0003] The problem to be solved by the present application is the automatic generation of document tags.

[0004] To solve the above problems, the present application adopts the following solutions:

[0005] A method for automatically generating document tags according to the present application comprises the following steps:

[0006] Step S1: receiving the summary information, author information and user-defined tag set of the target document, and obtaining a set of field tag trees; the field tag tree is a tree structure composed of field tag nodes; the field tag nodes correspond to fields, and the parent-child relationship between the field tag nodes represents the inclusion relationship between the corresponding fields;

[0007] The field tag nodes include a field name and a field tag set;

[0008] The field tag set is a keyword set composed of keywords synonymous with or related to the field name;

[0009] Step S2: based on the keywords in the field tag set in the set of field tag trees, performing word segmentation on the summary information to obtain a first tag set; at the same time, based on the author information of the document, obtaining a second tag set of the author's expert academic field; merging the first tag set, the second tag set and the user-defined tag set to obtain a third tag set;

[0010] Step S3: determining the corresponding field by matching the keywords in the field tag set in the set of field tag trees to obtain a document matching field set corresponding to the third tag set;

[0011] Step S4: analyzing the focus field of the document matching field set, and then determining a document focus field set according to the focus field;

[0012] Step S5: According to the document focus field set, a corresponding field label set is extracted from the field label tree set to form a first document label set, and then the self-defined label set is merged to obtain a final document label set.

[0013] Further, according to the document label automatic generation method of the present application, if the target document contains text content, the text content of the target document is also segmented when the summary information is segmented in step S2.

[0014] Further, according to the document label automatic generation method of the present application, step S4 includes the following steps:

[0015] Step SA41: According to the field label tree node relationship, the document matching field set is sorted in descending order of distance from the root node to obtain a document matching field sequence;

[0016] Step SA42: According to the sequence of the document matching field sequence, the fields are traversed, and according to the field label tree node relationship, the node corresponding to the traversed field is taken as the starting node, the nodes are gradually traversed to the root node, and through the number of keywords of the field containing the descendant fields in the document matching field set and itself in the third label set, if the number of keywords exceeds a certain proportion of the total number of keywords in the third label set, it is determined that the node is a focus field label node;

[0017] Step SA43: According to the field label tree node relationship, the fields belonging to the descendant nodes of the focus field label node are extracted from the document matching field set to form a document focus field set.

[0018] Further, according to the document label automatic generation method of the present application, in step S2, when segmenting, the word frequency is counted, and the word frequency proportion of each keyword in the first label set is calculated as the first relevance of the keyword; Each keyword in the first label set corresponds to a first relevance; When the first label set is merged with the second label set and the self-defined label set, the first relevance is merged in a weight addition manner, and then the third relevance corresponding to each keyword in the third label set is obtained after normalization; Step S4 includes the following steps:

[0019] Step SB41: According to the field label tree node relationship, the document matching field set is sorted in descending order of distance from the root node to obtain a document matching field sequence;

[0020] Step SB42: traversing the field in the order of the document matching field sequence, according to the field label tree node relationship, taking the node corresponding to the traversed field as the starting node, traversing the node to the root node step by step, and extracting the keywords corresponding to the fields in the third label set through the field containing the descendant fields in the document matching field set and itself, and counting the sum of the third relevance of the keywords as the document relevance of the field; if the document relevance value exceeds a certain threshold, the node is determined as the focus field label node;

[0021] Step SB43: according to the field label tree node relationship, extracting the fields belonging to the descendant nodes of the focus field label node from the document matching field set and itself to form a document focus field set.

[0022] According to the document label automatic generation device of the application, the following modules are included:

[0023] Module M1 is used for receiving the summary information, author information and user marked custom label set of a target document, and obtaining a field label tree set; the field label tree is a tree structure composed of field label nodes; the field label nodes correspond to fields, and the parent-child relationship between the field label nodes represents the containing relationship between the corresponding fields;

[0024] The field label node includes a field name and a field label set;

[0025] The field label set is a keyword set composed of keywords synonymous with or related to the field name;

[0026] Module M2 is used for performing word segmentation on the summary information based on the keywords of the field label set in the field label tree set to obtain a first label set; at the same time, a second label set of the author's expert academic field is obtained according to the author information of the document; and the first label set, the second label set and the custom label set are merged to obtain a third label set;

[0027] Module M3 is used for determining the corresponding field by matching the keywords of the field label set in the field label tree set to obtain a document matching field set corresponding to the third label set;

[0028] Module M4 is used for analyzing the focus field of the document matching field set, and further determining a document focus field set according to the focus field;

[0029] Module M5 is used for extracting the corresponding field label set from the field label tree set according to the document focus field set to combine a first document label set, and then merging the custom label set to obtain a final document label set.

[0030] Further, according to the document label automatic generation device, if the target document contains text content, the text content of the target document is also segmented when the summary information is segmented in the module M2.

[0031] Further, according to the document label automatic generation device, the module M4 comprises the following modules.

[0032] The module MA41 is configured to sort the document matching field set in descending order of distance from the root node according to the field label tree node relationship to obtain a document matching field sequence.

[0033] The module MA42 is configured to traverse the fields in the order of the document matching field sequence, and according to the field label tree node relationship, take the node corresponding to the traversed field as a starting node, traverse the nodes step by step towards the root node, and extract the keywords corresponding to the fields in the third label set from the document matching field set and the field itself, and count the sum of the third relevancies of these keywords as the document relevance of the field. If the document relevance value exceeds a certain threshold, the node is determined as a focus field label node.

[0034] The module MA43 is configured to extract the fields belonging to the descendant nodes of the focus field label node from the document matching field set to form a document focus field set according to the field label tree node relationship.

[0035] Further, according to the document label automatic generation device, when the module M2 segments, the word frequency is counted, and the proportion of the word frequency of each keyword in the first label set is calculated as the first relevance of the keyword. Each keyword in the first label set corresponds to a first relevance. When the first label set is merged with the second label set and the self-defined label set, the first relevance is merged in a weight addition manner, and then the third relevance of each keyword in the third label set obtained after normalization is obtained. The module M4 comprises the following modules.

[0036] The module MB41 is configured to sort the document matching field set in descending order of distance from the root node according to the field label tree node relationship to obtain a document matching field sequence.

[0037] Module MB42 is used for: traversing the fields in the order of the document matching field sequence, according to the field label tree node relationship, taking the node corresponding to the traversed field as a starting node, traversing the node to the root node step by step, and extracting the corresponding keywords in the third label set through the field containing the descendant fields in the document matching field set, and counting the sum of the third relevance of the keywords as the document relevance of the field; if the document relevance value exceeds a certain threshold, the node is determined as a focus field label node;

[0038] Module MB43 is used for: according to the field label tree node relationship, extracting the fields belonging to the descendant nodes of the focus field label node from the document matching field set and itself to form a document focus field set.

[0039] According to the machine readable medium of the present application, the program instruction set stored in the medium can be read by the machine, and when the program instruction set stored in the medium is loaded and executed by the machine, the document label automatic generation method described above can be realized.

[0040] According to the electronic device of the present application, the processor and the memory are connected, and the memory stores a program instruction set that can be read by the processor, and when the program instruction set stored in the memory is loaded and executed by the processor, the document label automatic generation method described above can be realized.

[0041] The technical effects of the present application are as follows:

[0042] 1. The present application can shield the interference of non-focus field keywords, accurately determine the content center of the document, and accurately determine the corresponding label.

[0043] 2. The final actual technical effect of the present application mainly depends on the construction of the input field label tree. As long as the field label tree is configured accurately, the determined label is also accurate. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is the flow chart of the label automatic generation process of the present application.

[0045] Figure 2 is the structure schematic diagram of the technical document retrieval system embodiment of the present application.

[0046] Figure 3 is an example of the field label tree in the textile field. DETAILED DESCRIPTION

[0047] The present application will be further described in detail below in combination with the drawings.

[0048] Figure 2A technical document retrieval system is exemplified in the present embodiment, which comprises a plurality of servers 100 and a plurality of clients 200, and each server 100 forms a server cluster. The servers 100 and the clients 200 are connected through a network. The server 100 is the electronic device as mentioned above, which comprises a connected processor 101 and a memory 102. The memory 102 comprises, but is not limited to, a random access memory, a disk, and the like, for storing data and a program instruction set, that is, the machine readable medium as mentioned above. The processor 101 is used to execute the program instruction set. The server 100 realizes the classified storage and retrieval service of the technical document by executing the program instruction set through the processor 101. The client 200 is used to submit the document to the server 100, or send a retrieval request to the server 100, and the like. The document label automatic generation method of the present embodiment is a process executed by the server 100 when the client 200 submits the technical document to the server 100. When the server 100 accepts the technical document submitted by the client 200, the server 100 generates the corresponding label for the technical document according to the document label automatic generation method of the present embodiment, and generates the corresponding inverted index with the label as the keyword, so as to facilitate the fast retrieval in the subsequent retrieval.

[0049] The input of the present embodiment comprises the target document itself, the summary information of the target document, the author information, the custom label set marked by the user, and the domain label tree set. The target document here refers to the document corresponding to the generated label. In addition, the method of the present application is based on the analysis of the text. If the target document is a multimedia format document such as picture, audio, video, etc., whether the target document itself is used as the input has no effect on the subsequent processing of the present application, that is, at this time, the target document itself is unnecessary.

[0050] Of course, in another preferred embodiment, the content of the multimedia format document such as picture, audio, video, etc. can be converted into text content by artificial intelligence, at this time, the converted text content can also be used as the input of the present application. There are many specific methods in the prior art for converting the content of the multimedia format document such as picture, audio, video, etc. into text content, and it is not the scope of the present application, and the present specification does not need to be repeated.

[0051] In the above input, the target document itself, the summary information of the target document, the author information, and the custom label set marked by the user are the contents input by the user through the client 200; and the domain label tree set is the system configuration input. The domain label tree set is a collection of domain label trees. The domain label tree is a tree structure composed of domain label nodes. The parent-child relationship between the domain label nodes represents the subordinate or containing relationship between the domains.

[0052] For example, referring to Figure 3The textile field includes the subfields of fiber materials, spinning, yarns, weaving, dyeing and finishing, and the like; the fiber materials include the subfields of plant fibers, animal fibers, organic chemical fibers, inorganic chemical fibers, and the like; the spinning field includes the subfields of fiber materials, spinning machines, finished yarns, and the like; the weaving field includes the subfields of flat weaving, three-dimensional weaving, knitting, and the like; the flat weaving includes the subfields of yarn materials, warping, sizing, looms, fabric structures, and the like; and the like.

[0053] In this embodiment, the field tag node includes a node ID, a field name, and a field tag set. The field tag set is a keyword set composed of keywords synonymous with or related to the field name. For example, the keywords synonymous with or related to yarn materials or yarns include yarn materials, yarns, twist, count, tex, breaking strength, breaking tenacity, and the like, and thus the field tag set of the yarn field tag node is a set composed of the keywords yarn materials, yarns, twist, count, tex, breaking strength, breaking tenacity, and the like.

[0054] In addition, to solve the problem of excessive space occupation of the set of field tag trees caused by field overlap and intersection, the field tag node in this embodiment introduces a cited node. For example Figure 3 In the example, the nodes represented by the dashed boxes, i.e., fiber materials, finished yarns, and yarn materials, are cited nodes. The cited node fiber materials essentially points to fiber materials. The cited nodes finished yarns and yarn materials essentially point to yarns.

[0055] The foregoing step S1 represents inputting the summary information, author information, user-labeled custom tag set, and field tag tree set of the target document. The document tag automatic generation method of the present application, with reference to Figure 1 , mainly includes a keyword extraction step, a keyword field matching step, a focus field analysis step, and a field-based tagging step.

[0056] The keyword extraction step is used to extract keywords based on the input. Specifically, with reference to the foregoing step S2, the keywords in the field tag set in the field tag tree set are used to perform word segmentation on the summary information to obtain a first tag set; at the same time, the second tag set of the author's expert academic field is obtained according to the author information of the document; and the first tag set, the second tag set, and the custom tag set are merged to obtain a third tag set. The first tag set, the second tag set, the custom tag set, and the third tag set are all keyword sets.

[0057] In another preferred embodiment, when performing word segmentation, if the target document itself contains text content or text content is extracted by artificial intelligence, the text content of the target document or the extracted text content is subjected to word segmentation at the same time.

[0058] In addition, it should be noted that the word segmentation is based on the keywords of the domain label set in the domain label tree set, that is, the word segmentation takes the keywords in the domain label tree set as the word source, so that the keyword set obtained by the word segmentation can be matched in the domain label tree set.

[0059] In this embodiment, the server 100 is equipped with an expert information library. The expert information library defines experts and corresponding academic fields. In step S2, the second label set of the author's expert academic field is obtained according to the author information of the document, that is, the author is searched in the expert information library, and the corresponding academic field is extracted. The corresponding academic field is matched with the domain in the domain label tree set to obtain the corresponding domain label node, and then the domain label set of the corresponding domain label node is combined into the second label set.

[0060] The keyword domain matching step is used to match the relevant domain according to the keywords. This step is the aforementioned step S3, that is, the third label set corresponding to the document matching domain set is obtained by matching the keywords of the domain label set in the domain label tree set. The document matching domain set is a set of domains, which is also a set of domain label nodes.

[0061] The focus domain analysis step, that is, the aforementioned step S4, analyzes the focus domain of the document matching domain set, and further determines the document focus domain set according to the focus domain. In a simple implementation, the focus domain can be obtained according to the distribution of the keywords in the third label set. In this simple implementation, step S4 includes the following steps:

[0062] Step SA41: According to the relationship between the domain label tree nodes, the document matching domain set is sorted in descending order of distance from the root node to obtain a document matching domain sequence;

[0063] Step SA42: Traverse the domains in the order of the document matching domain sequence, and according to the relationship between the domain label tree nodes, take the node corresponding to the traversed domain as the starting node, gradually traverse the nodes to the root node, and through the number of keywords in the third label set that the domain contains the descendant domains in the document matching domain set and itself, if the number of keywords exceeds a certain percentage of the total number of keywords in the third label set, it is determined that the node is a focus domain label node;

[0064] Step SA43: According to the relationship between the domain label tree nodes, extract the domains of the descendant nodes belonging to the focus domain label node and itself from the document matching domain set to form the document focus domain set.

[0065] In step SA42, the certain percentage of the total number of keywords is generally set to 75-90%, and the preferred value in this embodiment is 80%.

[0066] Taking the textile field label tree as an example, suppose that the keywords in the third label set obtained after a certain document is processed by the foregoing step S2 are as follows:

[0067] pure cotton, warp density, fabric, warp, weft, eight warp surface, satin weave, satin surface fabric, linear density, breaking strength, combed cotton yarn, warping, sizing, reeding, weaving, dyeing and finishing, single shaft share, warping beam, yarn, sizing tank, double dipping and double pressing, sizing beam, air jet loom, heald, heald wire.

[0068] The total number of keywords in the third label set is 25.

[0069] After the step S3 is processed, the fields in the document matching field set obtained are as follows:

[0070] fiber material, fiber material, yarn, finished yarn, yarn material, flat weaving, dyeing and finishing, flat weaving, warping, sizing, loom, fabric organization.

[0071] After the step SA41 is sorted, the document matching field sequence obtained is as follows:

[0072] yarn material, warping, sizing, loom, fabric organization, fiber material, finished yarn, flat weaving, fiber material, yarn, flat weaving, dyeing and finishing.

[0073] In step SA42, the parent node of yarn material, warping, sizing, loom, and fabric organization is flat weaving, and the corresponding descendant fields of the document matching field set are yarn material, warping, sizing, loom, and fabric organization. The keywords corresponding to these fields in the third label set include:

[0074] pure cotton, warp density, fabric, warp, weft, eight warp surface, satin weave, satin surface fabric, linear density, breaking strength, combed cotton yarn, warping, sizing, reeding, weaving, single shaft share, warping beam, yarn, sizing tank, double dipping and double pressing, sizing beam, air jet loom, heald, heald wire.

[0075] The number of keywords is 24. Compared with the total number of keywords in the third label set, which is 25, the proportion exceeds 80%, so the focus field label node is flat weaving. Then, by step SA43, the document focus field set obtained by extracting flat weaving and flat weaving child nodes from the document matching field set is as follows:

[0076] yarn material, warping, sizing, loom, fabric organization, flat weaving.

[0077] It should be noted that in step SA43, if the field corresponding to the focus field tag node is in the third label set, the document focus field set contains the focus field tag node; if the field corresponding to the focus field tag node is not in the third label set, the document focus field set does not contain the focus field tag node. For example, in the aforementioned example, the focus field node plane is in the third label set, and the field tag node in the document focus field set includes the plane weave.

[0078] In another complex implementation, the focus field can be obtained according to the distribution of the keyword frequency proportion in the field in the third label set. In this implementation, it is required to count the word frequency when performing word segmentation in the aforementioned step S2, and to calculate the word frequency proportion. Specifically, when performing word segmentation, the word frequency is counted, and the word frequency proportion of each keyword in the first label set is calculated as the first relevance of the keyword. Each keyword in the first label set corresponds to a first relevance. When the first label set is merged with the second label set and the custom label set, the first relevance is merged in a weight addition manner, and then the third relevance corresponding to each keyword in the third label set is obtained after normalization. In this implementation, step S2 can be decomposed into the following steps:

[0079] Step S21, based on the keywords of the field label set in the field label tree set, the summary information and / or the text content are segmented, and the word frequency of each keyword is counted, to obtain a first label set;

[0080] Step S22, calculate the first relevance of each word frequency in the first label set: first relevance = C_i / C_total; wherein C_i represents the word frequency of the i-th keyword in the first label set, and C_total represents the total of all keyword frequencies in the first label set;

[0081] Step S23, obtain the second label set of the author's expert academic field according to the author information of the document;

[0082] Step S24, merge the first label set, the second label set and the custom label set to obtain the third label set, and merge the first relevance in a weight addition manner to calculate the third relevance, the specific steps are as follows:

[0083] Step S241: the third relevance is initialized to 0;

[0084] Step S242: if the keyword is in the first label set, the third relevance is added by a1*first relevance;

[0085] Step S243: if the keyword is in the second label set, the third relevance is added by a2 / C_count2;

[0086] Step S244: If the keyword is in the self-defined label set, the third correlation degree is accumulated a3 / C_count3;

[0087] wherein a1, a2, a3 are weight coefficients, C_count2 is the total number of keywords in the second label set, and C_count3 is the total number of keywords in the self-defined label set.

[0088] Step S25: The third correlation degrees are normalized so that the third correlation degree values of the keywords in the third label set add up to 1.

[0089] In the above embodiment, step S4 comprises the following steps:

[0090] Step SB41: According to the relationship between the domain label tree nodes, the document matching domain set is sorted in descending order of distance from the root node to obtain a document matching domain sequence.

[0091] Step SB42: The domains are traversed in the order of the document matching domain sequence, and according to the relationship between the domain label tree nodes, the nodes are traversed step by step from the node corresponding to the traversed domain as the starting node to the root node, and the keywords corresponding to the nodes are extracted in the third label set through the domain containing the descendant domains in the document matching domain set and itself, and the sum of the third correlation degrees of these keywords is taken as the document correlation degree of the domain; if the document correlation degree value exceeds a certain threshold value, the node is determined as the focus domain label node.

[0092] Step SB43: According to the relationship between the domain label tree nodes, the domains belonging to the descendant nodes of the focus domain label node are extracted from the document matching domain set to form a document focus domain set.

[0093] In step SB42, the threshold value of the document correlation degree value is generally set to 0.7-0.9, and the preferred value in this embodiment is 0.8.

[0094] Compared with the foregoing simple embodiment, the processing process of this embodiment is basically the same as that of the foregoing simple embodiment, and the difference lies in that the former takes the number of keywords as the statistical object, and the latter takes the third correlation degree and as the statistical object. For example, in the foregoing example of the third label set, the third correlation degree of each keyword is 0.04. In step SB42, the parent node of yarn material, warping, sizing, loom, and fabric organization is plane weaving, and the descendant domains corresponding to the domain label node in the document matching domain set are yarn material, warping, sizing, loom, and fabric organization. The sum of the third correlation degrees of the 24 keywords in the third label set of these domains plus the domain itself is 0.96, which exceeds the threshold value 0.8, so the focus domain label node is plane weaving.

[0095] Based on the field tagging step, that is, the aforementioned step S5, according to the document focus field set, a corresponding field tag set is extracted from the field tag tree set to form a first document tag set, and then the custom tag set is merged to obtain a final document tag set. In the foregoing example, assume that the document focus field set is {yarn material, warping, sizing, loom, fabric organization, flat weaving}, and then the field tag sets of the yarn material, warping, sizing, loom, fabric organization, and flat weaving fields are extracted from the field tag tree set to form a first document tag set. After the first document tag set and the custom tag set are merged, a final document tag set is obtained.

[0096] In addition, it also needs to be explained that the device and the method in the application correspond to a virtual device, and the modules in the device correspond to the steps in the method, which will not be repeated.

Claims

1. A method for automatic generation of document tags, characterized by, The method comprises the following steps: Step S1: receiving summary information, author information and a set of user-labeled custom tags of a target document, and obtaining a set of domain tag trees; the domain tag tree is a tree structure composed of domain tag nodes; the parent-child relationship between the domain tag nodes represents the containing relationship between the corresponding domains; The domain tag node comprises a domain name and a set of domain tags; The set of domain tags is a set of keywords composed of keywords synonymous with or related to the domain name; Step S2: performing word segmentation on the summary information based on the keywords in the set of domain tags in the set of domain tag trees to obtain a first set of tags; simultaneously, obtaining a second set of tags of the academic domain of the author of the document according to the author information of the document; merging the first set of tags, the second set of tags and the set of custom tags to obtain a third set of tags; Step S3: determining the corresponding domain by matching the keywords in the set of domain tags in the set of domain tag trees to obtain a set of document matching domains corresponding to the third set of tags; Step S4: analyzing the focus domain of the set of document matching domains, and then determining a set of document focus domains according to the focus domain; Step S5: according to the set of document focus domains, extracting the corresponding set of domain tags from the set of domain tag trees to combine into a first set of document tags, and then merging the set of custom tags to obtain a final set of document tags.

2. The method of claim 1, wherein, If the target document contains text content, the summary information is segmented in step S2, and the text content of the target document is also segmented.

3. The method of claim 1 or 2, wherein The step S4 comprises the following steps: Step SA41: sorting the set of document matching domains in descending order of distance from the root node according to the relationship between the domain tag nodes to obtain a document matching domain sequence; Step SA42: traversing the domains in the order of the document matching domain sequence, taking the node corresponding to the traversed domain as the starting node, and gradually traversing the nodes to the root node according to the relationship between the domain tag nodes, and through the number of keywords in the set of document matching domains and the number of keywords in the third set of tags, if the number of keywords exceeds a certain proportion of the total number of keywords in the third set of tags, the node is determined as a focus domain tag node; Step SA43: extracting the domains of the descendant nodes belonging to the focus domain tag node and itself from the set of document matching domains to form a set of document focus domains according to the relationship between the domain tag nodes.

4. The method of claim 1 or 2, wherein In step S2, when performing word segmentation, the word frequency is counted, and the proportion of the word frequency of each keyword in the first set of tags is calculated as the first relevance of the keyword; each keyword in the first set of tags corresponds to a first relevance; when the first set of tags is merged with the second set of tags and the set of custom tags, the first relevance is merged in the form of weight addition, and then the third relevance of each keyword in the third set of tags obtained after normalization is obtained; the step S4 comprises the following steps: Step SB41: according to the field tag tree node relationship, the document matching field set is sorted in descending order of distance from the root node to obtain a document matching field sequence; Step SB42: according to the field tag tree node relationship, the field is traversed in the order of the document matching field sequence, and the node corresponding to the traversed field is taken as the starting node, the node is gradually traversed to the root node, and the keywords corresponding to the field are extracted from the document matching field set and the third label set, and the sum of the third relevance of these keywords is taken as the document relevance of the field; if the document relevance value exceeds a certain threshold, the node is determined as the focus field tag node; Step SB43: according to the field tag tree node relationship, the fields belonging to the descendant nodes of the focus field tag node are extracted from the document matching field set to form a document focus field set.

5. An apparatus for automatic generation of document tags, characterized by The method comprises the following steps: Module M1 is configured to receive the summary information, author information and user marked custom label set of the target document, and obtain a field label tree set; the field label tree is a tree structure composed of field label nodes; the parent-child relationship between the field label nodes represents the containing relationship between the corresponding fields; The field label node comprises a field name and a field label set; The field label set is a keyword set composed of keywords synonymous with or related to the field name; Module M2 is configured to perform word segmentation on the summary information based on the keywords of the field label set in the field label tree set to obtain a first label set; at the same time, a second label set of the author's expert academic field is obtained according to the author information of the document; the first label set, the second label set and the custom label set are merged to obtain a third label set; Module M3 is configured to determine the corresponding field by matching the keywords of the field label set in the field label tree set to obtain a document matching field set corresponding to the third label set; Module M4 is configured to analyze the focus field of the document matching field set, and then determine a document focus field set according to the focus field; Module M5 is configured to extract the corresponding field label set from the field label tree set according to the document focus field set to combine a first document label set, and then merge the custom label set to obtain a final document label set.

6. The apparatus for automatic generation of document tags as recited in claim 5, wherein, If the target document contains text content, the module M2 also performs word segmentation on the text content of the target document when performing word segmentation on the summary information.

7. The apparatus for automatic generation of document tags according to claim 5 or 6, wherein, The module M4 comprises the following modules: Module MA41 is configured to sort the document matching field set in descending order of distance from the root node according to the field tag tree node relationship to obtain a document matching field sequence; Module MA42 is configured to traverse the fields in the order of the field matching domain sequence, and according to the field label tree node relationship, to take the node corresponding to the traversed field as a starting node, to traverse the node gradually to the root node, and to extract the keywords corresponding to the field in the third label set through the field containing the descendant fields in the document matching domain set and the field itself, to count the sum of the third relevancies of the keywords as the document relevance of the field; and to determine the node as a focus field label node if the document relevance value exceeds a certain threshold. Module MA43 is configured to extract the fields belonging to the descendant nodes of the focus field label node from the document matching domain set to form a document focus domain set.

8. The apparatus for automatically generating a document tag according to claim 5 or 6, wherein, In the module M2, when performing word segmentation, the word frequency is counted, and the word frequency proportion of each keyword in the first label set is calculated as the first relevance of the keyword; each keyword in the first label set corresponds to a first relevance; when the first label set is merged with the second label set and the self-defined label set, the first relevance is merged in a weight addition manner, and then the third relevance of each keyword in the third label set obtained after normalization is obtained; the module M4 includes the following modules: Module MB41 is configured to sort the document matching domain set in descending order of distance from the root node to obtain a document matching domain sequence according to the field label tree node relationship; Module MB42 is configured to traverse the fields in the order of the document matching domain sequence, and according to the field label tree node relationship, to take the node corresponding to the traversed field as a starting node, to traverse the node gradually to the root node, and to extract the keywords corresponding to the field in the third label set through the field containing the descendant fields in the document matching domain set; and to determine the node as a focus field label node if the document relevance value exceeds a certain threshold. Module MB43 is configured to extract the fields belonging to the descendant nodes of the focus field label node from the document matching domain set to form a document focus domain set.

9. A machine-readable medium having stored therein a set of program instructions executable by a machine, the set of program instructions comprising: When the program instruction set stored in the medium is loaded and executed by the machine, the method for automatically generating document labels according to any one of claims 1 to 4 can be implemented.

10. An electronic device comprising a processor and a memory connected thereto, said memory having stored therein a set of program instructions readable by the processor, characterized in that, When the program instruction set stored in the memory is loaded and executed by the processor, the method for automatically generating document labels according to any one of claims 1 to 4 can be implemented.

Images