Digital library index retrieval method and system

Abstract

The invention discloses a digital library index retrieval method and system, and relates to the technical field of natural language process.The method comprises the steps that a query statement input by a user is received, the semantic offset degree of the query statement in a cross-language environment is evaluated based on a multi-language semantic mapping mechanism, and a corresponding retrieval risk level is determined according to the semantic offset degree; a query statement is complemented and enhanced by introducing a multi-language semantic anchor point and concept association relationship; and performing semantic matching and scoring according to the compensated query content and the target literature, and executing a dynamic index decision operation based on a set risk level and an integrity score. According to the method, the multi-language semantic anchor point diagram is constructed, the edge weight path scoring function is combined, accurate and credible evaluation is carried out on the semantic compensation path, and the compensation result is more stable and controllable. Through semantic anchor point path extension and a multi-hop semantic propagation mechanism, original query semantic expression is complemented, and the perception ability of a multi-language retrieval system for weak explicit semantics is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, and in particular to a digital library index retrieval method and system. BACKGROUND

[0002] In recent years, with the rapid growth of global digital libraries and cross-language resources, when users query complex semantic content (such as "company acquisition of start-ups") in Chinese, English or other languages, the traditional methods based on keyword matching or shallow vector semantics often fail to accurately retrieve semantically related but differently expressed literature resources. This mainly manifests in the following two aspects: semantic drift problem: the words in the query text may drift in semantics in different languages or knowledge domains, that is, the concept of the original word changes slightly but critically in the target language or context, causing the index system to fail to correctly match the literature. For example, "acquisition" in Chinese may mean "acquisition" or "takeover" or "buyout" in English, and even "merge" in some technical fields. Semantic loss and word meaning loss problem: users may use partially incomplete semantic expressions or have ambiguous spaces in expression habits when inputting queries, causing the system to ignore superordinate and subordinate concepts, equivalent translation nodes or domain semantic anchors when constructing semantic representations, thereby reducing the precision and recall rate of retrieval.

[0003] Some existing retrieval methods attempt to use word vectors, bilingual dictionaries or cross-language retrieval models based on attention mechanisms to compensate, but have the following limitations: the degree of word drift in the cross-language semantic mapping process is not considered; after semantic drift occurs, the compensation mechanism is not automatically triggered to restore the complete semantic intent; in the semantic compensation process, structured semantic anchor networks (SAN) are not constructed using domain priors, so there is a lack of stable and reliable compensation basis; the syntactic edge set information is not effectively used to assist in identifying semantic cores after the dependency syntax tree is constructed. SUMMARY

[0004] This section is intended to summarize some aspects of the embodiments of the present application and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of the specification to avoid obscuring the purpose of this section, the abstract and the title, and such simplifications or omissions cannot be used to limit the scope of the present application.

[0005] In view of the problems in the prior art, the present application is proposed.

[0006] To solve the above technical problems, the present application provides the following technical solutions: a digital library index retrieval method, the method comprising the following steps:

[0007] S1. receiving a query sentence input by a user, and performing language structure analysis on the query sentence to extract key semantic components;

[0008] S2. based on a multi-language semantic mapping mechanism, evaluating the semantic deviation degree of the query sentence in a cross-language environment, and determining the corresponding retrieval risk level accordingly;

[0009] S3. when detecting that the semantic deviation exceeds a preset threshold, activating a semantic compensation mechanism, and completing and enhancing the query sentence by introducing multi-language semantic anchor points and concept association relationships;

[0010] S4. performing semantic matching and scoring according to the compensated query content and the target literature, and performing dynamic indexing decision operation based on the set risk level and integrity score.

[0011] As a preferred scheme of the digital library index retrieval method, the extraction method of the key semantic components is: through the structured processing of the query text by the dependency syntax analyzer, a dependency syntax tree T=(V, E) is constructed, wherein V represents all word nodes in the text, that is, a word node set; E represents the syntax dependency relationship between nodes, that is, a syntax relationship edge set;

[0012] The word node set is filtered based on the node part of speech, and based on the syntax relationship edge set E, the word nodes in the main path or high dependency weight position in the syntax structure are selected as core words.

[0013] As a preferred scheme of the digital library index retrieval method, the evaluation method of the semantic deviation degree is:

[0014] S201: a pre-trained cross-language word vector mapping matrix is introduced, which converts the word vector in the query language (L1) into a vector form in the target language (L2) space through the mapping, and the vector mapping matrix is represented as: M L1→L2 , and the function is:

[0015] S202: for each core word v i , the mapping result of its semantics in the target language space is: M L1→L2 (v i ), taking the average semantic reference vector in the target language as a reference point, and taking its semantic distribution standard deviation σ L2 for normalization processing to obtain the normalized drift of the word in the language space;

[0016] S203: based on the dependency syntax edge set E, the syntax importance weight w i of each word node is constructed;

[0017] S204: Based on the weight w i The standardization semantic drift amount of all core words is weighted and averaged to obtain the overall semantic drift degree of the entire query text after language conversion.

[0018] As a preferred scheme of the digital library index retrieval method, the retrieval risk level is divided into multiple risk intervals R based on the continuous numerical range of the semantic drift degree δ, including but not limited to: a semantic stability low-risk interval, a slight drift medium-low risk interval, a moderate drift medium risk interval, an obvious deviation medium-high risk interval, and a high-risk ambiguity high-risk interval.

[0019] As a preferred scheme of the digital library index retrieval method, the semantic compensation mechanism is:

[0020] A semantic anchor point network SAN is constructed, which includes a multilingual concept node set A and a concept relationship edge set R. When δ> δ t If δ> δ, it is determined that there is a serious semantic deviation risk, the SAN network is activated, and the semantic anchor compensation logic is started for the word.

[0021] S301: For the triggered center word, the relevant nodes and their semantic adjacent nodes are retrieved from the SAN;

[0022] S302: Based on the edge set R, a graph structure sub-path tree with the center word as the source point is constructed;

[0023] S303: According to the score of each candidate path multiplied by the node semantic similarity, the final candidate weight is calculated, the top K candidate anchors are reserved, and the final compensation word set is formed;

[0024] S304: The selected anchor words are embedded in the original task in a parallel form, an annotation form, or a semantic weighted form, and the query sentence is completed.

[0025] As a preferred scheme of the digital library index retrieval method, the method for semantic matching and scoring in the S4 process is:

[0026] S401: The anchor set coincidence degree is used to evaluate whether the candidate literature directly contains the anchor points in the query sentence;

[0027] S402: A compensation intensity coefficient is introduced Where k represents the attenuation factor, δ t When the semantic deviation exceeds the preset threshold, the anchor semantic approximation compensation term is established to evaluate the semantic similarity between the anchor point pairs;

[0028] S403: The final score function is constructed by comprehensively considering the set coincidence degree of anchor points and the semantic approximation compensation term of anchor points: S = alpha * set coincidence degree of anchor points + beta * semantic approximation compensation term of anchor points, and the score is calculated through the function.

[0029] As a preferred scheme of the digital library index retrieval method, the index decision operation comprises:

[0030] If the risk level is in the moderate drift medium risk interval or the obvious deviation high risk interval, and the score S is less than a threshold S1, cross-language index incremental training is triggered.

[0031] If the risk level is in the high risk ambiguity high risk interval or S is less than a threshold S2, an artificial review process is started.

[0032] The retrieval system applied to the above-mentioned digital library index retrieval method comprises:

[0033] The query analysis module receives the query text input by the user, performs language analysis on the query content, including dependency syntax analysis and keyword extraction; the semantic deviation evaluation module calculates the deviation degree of the core word in the multilingual semantic space by using a cross-language mapping model, and outputs a discretized drift risk level;

[0034] The risk-driven compensation module determines whether to trigger a semantic anchor compensation operation according to the risk level, generates and adjusts a compensation coefficient, and establishes a semantic anchor network; the semantic matching score module calculates a semantic integrity score based on the anchor network and the semantic content of the query literature, which is used to judge the relevance and completeness of the retrieval result; the decision control module makes a response operation according to the risk level and the score result, such as incremental training, artificial review or default passing.

[0035] The application further discloses a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the digital library index retrieval method when executing the computer program.

[0036] The application further discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the digital library index retrieval method.

[0037] The application has the following beneficial effects:

[0038] 1. The semantic drift amount computer mechanism is effective: the application quantitatively evaluates the semantic drift degree in different language expressions by using a normalized cross-language semantic vector deviation formula, and improves the sensitive detection ability of query intention error;

[0039] 2. Structured semantic anchor point network path completion path credibility scoring mechanism: by constructing a multilingual semantic anchor point graph, combining edge weight path scoring functions, and accurately and reliably evaluating semantic compensation paths, the compensation results are more stable and controllable. Through the semantic anchor point path expansion and multi-hop semantic propagation mechanism, the original query semantic expression is completed, and the perception ability of the multilingual retrieval system to weak explicit semantics and the literature matching rate are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0041] Figure 1 The overall flow principle schematic diagram of a digital library index retrieval method proposed by the present application. DETAILED DESCRIPTION

[0042] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0043] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0044] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0045] REFERENCE Figure 1 For one embodiment of the present application, a digital library index retrieval method and system are provided, which includes the following steps:

[0046] S1. Receiving a user input query sentence and performing language structure analysis on the query sentence to extract key semantic components; the extraction method of key semantic components is: performing structured processing on the query text by a dependency syntax analyzer to construct a dependency syntax tree T=(V,E), wherein V represents all word nodes in the text, i.e. word node set; E represents the syntactic dependency relationship (such as subject-predicate, verb-object, and modifier) between nodes, i.e. the syntax relationship edge set;

[0047] The word node set is filtered based on the part of speech of the node (such as noun, verb, etc.), and based on the syntactic relationship edge set E, the word nodes in the main path or high dependency weight position in the syntactic structure are selected as the core words.

[0048] Example: Query text (Chinese): "The company acquired the start-up", the dependency syntactic tree is constructed: node set V = {company, acquisition, start-up}; syntactic edge set E: "company" → "acquisition" (subject-predicate), "acquisition" → "start-up" (verb-object), extract core words: since "acquisition" is the main predicate of the sentence, "company" and "start-up" are the subject and object, all three are high semantic contribution words, and are selected as core words.

[0049] S2. Based on the multilingual semantic mapping mechanism, the semantic shift degree of the query sentence in the cross-language environment is evaluated, and the corresponding retrieval risk level is determined accordingly;

[0050] Specifically, the evaluation method of the semantic shift degree is:

[0051] S201: In order to measure the semantic transfer difference between languages, a pre-trained cross-language word vector mapping matrix is introduced, which converts the word vector in the query language (L1) into a vector form in the target language (L2) space through the mapping. The vector mapping matrix is represented as: M L1→L2 Its function is:

[0052] S202: For each core word v i , the mapping result of its semantics in the target language space is: M L1→L2 (v i ), taking the average semantic reference vector in the target language as the reference point, and taking its semantic distribution standard deviation σ L2 for normalization, to obtain the normalized drift of the word in the language space;

[0053] Its calculation formula can be expressed as:

[0054] The meaning is to eliminate the inconsistency of the scale in different semantic spaces; to ensure the comparability and stability of the drift measure in different language mappings;

[0055] S203: Based on the dependency syntax edge set E, the syntactic importance weight w i of each word node is constructed.

[0056] The calculation method of w i may combine the following indicators:

[0057] 1. The path length between the word and the central predicate;

[0058] 2. The number of out-edges of the word as a dependency center (i.e. the grammatical scope of influence on other words);

[0059] 3. The grammatical type of the connected edge (such as subject-predicate > modifier > parallel, giving different weights to different edges).

[0060] For example: the weight of the subject or object should be higher than that of the modifier or attributive word.

[0061] By introducing w i , the direct integration of syntactic structure in semantic drift evaluation is realized, so that the calculation result not only depends on the word vector space, but also integrates the grammatical skeleton in the language structure.

[0062] S204: Weighted average of the standardized semantic drift of all core words based on the weight w i , to obtain the overall semantic drift degree of the entire query text after language conversion.

[0063] The calculation formula can be:

[0064] Where n represents the number of core words, and δ represents the overall semantic drift degree. The larger the value, the stronger the semantic drift in the target language.

[0065] In addition, the retrieval risk level is divided into multiple risk intervals R based on the continuous numerical range of the semantic drift degree δ, including but not limited to: a semantic stability low-risk interval, a slight drift medium-low risk interval, a moderate drift medium risk interval, a significant deviation medium-high risk interval, and a high-risk ambiguity high-risk interval.

[0066] S3. When the semantic drift is detected to exceed the preset threshold, activate the semantic compensation mechanism, and complete and enhance the query sentence by introducing multilingual semantic anchor points and concept association relationships;

[0067] Specifically, the semantic compensation mechanism is:

[0068] Build a semantic anchor point network SAN, which includes a set of multilingual concept nodes A (composed of a plurality of synonymous words, hyponyms, derivatives, etc. in different languages, forming an anchor node set) and a set of concept relationship edges R (the edge set R represents the semantic relationship between word anchors). After satisfying δ t > δ (Pre-set threshold), it is determined that there is a serious semantic drift risk, and the SAN network is activated to start the semantic anchor compensation logic for the word, the method is:

[0069] S301: For the triggered central word, retrieve the relevant nodes and their semantic adjacent nodes from SAN;

[0070] denoted as: A = {a1, a2,..., a m}|d(v i ,a j )≤T d , where d(v i ,a j ) is semantic distance (such as word vector Euclidean distance or graph structure hop count), T d is the maximum semantic expansion range, nodes may come from different languages, but have semantic alignment properties.

[0071] S302: Based on the edge set R, construct a graph structure sub-path tree with the center word as the source point;

[0072] denoted as: Each path will be calculated score represented as: Π r∈P w r , edge type weight w r may be predefined, such as: the most credible synonym relationship w r take 1; Context relationship second w r take 0.7; Abstract class meaning is weak w r take 0.5.

[0073] S303: According to the score of each candidate path multiplied by the node semantic similarity (calculated by sim() similarity function), calculate the final candidate weight: sim(v i ,a j )·Π r∈P w r , keep the top K candidate anchor points, form the final compensation word set;

[0074] S304: Embed the selected anchor words into the original task in parallel form, annotation form or semantic weighting form, complete the completion of the query sentence

[0075] S4. According to the compensation query content and the target literature, perform semantic matching and scoring, and based on the set risk level and integrity score, perform dynamic index decision operation.

[0076] Specifically, the method for semantic matching and scoring in S4 process is:

[0077] S401: Evaluate whether the candidate literature directly contains the anchor points in the query sentence by establishing the anchor point set coincidence degree;

[0078] The expression of establishing the anchor point set coincidence degree is: The higher the value, the more consistent the literature and the query semantic expression;

[0079] S402: Introduce compensation strength coefficient where k denotes the decay factor, δ t where the semantic shift exceeds the preset threshold, an anchor point semantic approximation compensation term is established to evaluate the semantic similarity between anchor point pairs.

[0080] The expression of the anchor point semantic approximation compensation term is: where m denotes the total number of semantic anchors, v Qi denotes the embedding vector representation of the i-th semantic anchor in the query text in the semantic anchor network (SAN), v Di denotes the embedding vector representation of the anchor point in the SAN that matches or is closest to the i-th query anchor in the candidate document, and sim(·) is a similarity function.

[0081] S403: The final scoring function S is constructed by integrating the anchor set coincidence degree and the anchor point semantic approximation compensation term: S = a anchor set coincidence degree + β anchor point semantic approximation compensation term, and the score is calculated by the function

[0082] In addition, it should be noted that the index decision operation includes:

[0083] If the risk level is in the moderate drift medium risk interval or the obvious deviation high risk interval (indicating significant semantic drift), and the score S is less than the threshold S1 (for example, S1 takes a value range of [0.4, 0.6], indicating that it cannot support high-quality matching), trigger the cross-language index incremental training; including: updating the cross-language mapping matrix M L1→L2 , and adjusting the inverted index structure to improve the semantic alignment capability between subsequent queries and documents

[0084] If the risk level is in the high-risk ambiguity high-risk interval or S is less than the threshold S2 (for example, S2 takes a value range of [0.2, 0.4), indicating that the score is significantly too low, even if the semantic drift is not obvious, it also indicates that the SAN is insufficient for this type of query coverage), start the manual review process.

[0085] The embodiment also provides a digital library index retrieval system, according to the digital library index retrieval method of claim 7, the system comprises:

[0086] A query analysis module receives user input query text, and performs language analysis on the query content, including dependency syntax analysis and keyword extraction; a semantic shift evaluation module uses a cross-language mapping model to calculate the shift degree of core words in a multi-language semantic space, and outputs a discretized drift risk level;

[0087] The risk-driven compensation module determines whether to trigger a semantic anchor compensation operation according to the risk level, generates and adjusts a compensation coefficient, and establishes a semantic anchor network; the semantic matching scoring module calculates a semantic integrity score based on the anchor network and the semantic content of the query literature, and the semantic integrity score is used to determine the relevance and completeness of the search results.

[0088] The decision control module makes a response operation according to the risk level and the score result, such as incremental training, manual review or default passing.

[0089] The embodiment also provides a computer device suitable for the case of the digital library index retrieval method, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the digital library index retrieval method as described in the above embodiment.

[0090] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can also be an external keyboard, touchpad or mouse, etc.

[0091] The embodiment also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method for searching a digital library index according to the above embodiment. The storage medium can be implemented by any type of volatile or nonvolatile storage devices or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or a compact disk.

[0092] It should be noted that the above embodiment is only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method of retrieving an index of a digital library, characterized by, The method comprises the following steps: S1. Receiving a user input query sentence and performing language structure analysis on the query sentence to extract key semantic components; S2. Based on a multi-language semantic mapping mechanism, evaluating the semantic shift degree of the query sentence in a cross-language environment and determining the corresponding retrieval risk level accordingly; S3. When detecting that the semantic shift exceeds a preset threshold, activating a semantic compensation mechanism, and completing and enhancing the query sentence by introducing multi-language semantic anchor points and concept association relationships; S4. According to the compensated query content and the target document, performing semantic matching and scoring, and based on the set risk level and integrity score, performing dynamic index decision operation.

2. The method of claim 1, wherein: The extraction method of the key semantic components is: through a dependency syntax analyzer, the query text is structured and processed to construct a dependency syntax tree T=(V, E), wherein V represents all word nodes in the text, that is, a word node set; E represents the syntactic dependency relationship between nodes, that is, a syntax relationship edge set; The word node set is filtered based on the node part-of-speech, and based on the syntax relationship edge set E, the word nodes in the main path or high dependency weight position in the syntax structure are selected as the core words.

3. The method of claim 2, wherein: The evaluation method of the semantic shift degree is: S201: introduce a pre-trained cross-language word vector mapping matrix, convert the word vector in the query language into a vector form in the target language space through the mapping, and the vector mapping matrix is represented as: M L1→L2 The function is: S202: For each core word v i , the mapping result of its semantics in the target language space is: M L1→L2 (v i ), taking the average semantic reference vector in the target language as the reference point, and normalizing it with its semantic distribution standard deviation σ L2 to obtain the normalized drift of the word in the language space; S203: Construct the syntax importance weight w of each word node based on the dependency syntax side set E i ; S204: Weighted average of the normalized semantic shift of all core words based on the weight w i The overall semantic shift degree of the entire query text after language conversion is obtained by weighted average of the normalized semantic shift of all core words.

4. The method of claim 3, wherein: The retrieval risk level is divided into multiple risk intervals R based on the continuous numerical range of the semantic drift degree δ, including but not limited to: a semantic stability low-risk interval, a slight drift medium-low risk interval, a moderate drift medium risk interval, an obvious shift medium-high risk interval, and a high risk ambiguity high risk interval.

5. The method of claim 4, wherein: The semantic compensation mechanism is: A semantic anchor network SAN is constructed, which includes a multilingual concept node set A and a concept relation edge set R. When δ > δ t After that, it is determined that there is a serious semantic deviation risk, the SAN network is activated, and semantic anchor compensation logic is started for the word. S301: For the triggered center word, retrieve the related nodes and their semantic adjacent nodes from SAN; S302: Based on the edge set R, construct a graph structure sub-path tree with the center word as the source point; S303: Calculate the final candidate weight according to the score of each candidate path multiplied by the node semantic similarity, retain the top K candidate anchor points, and form the final compensation word set; S304: Embed the selected anchor words into the original task in a parallel form, an annotation form or a semantic weighted form to complete the completion of the query sentence.

6. The method of claim 5, wherein: The method for performing semantic matching and scoring in the S4 process is: S401: Evaluate whether the candidate document directly contains the anchor points in the query sentence by establishing the anchor point set overlap degree; S402: Introducing compensation intensity coefficient wherein k represents an attenuation factor, δ t The semantic offset exceeds the preset threshold, and the anchor point semantic approximation compensation term is established to evaluate the semantic similarity between the anchor point pairs. S403: Construct the final scoring function S=α·anchor point set overlap degree+β·anchor point semantic approximation compensation term by comprehensively considering the anchor point set overlap degree and the anchor point semantic approximation compensation term, and calculate the score through the function.

7. A method for retrieving an index of a digital library according to claim 6, characterized in that: The index decision operation includes: If the risk level is in the moderate drift medium risk interval or the obvious shift medium-high risk interval, and the score S is less than the threshold S1, trigger the cross-language index incremental training; If the risk level is in the high risk ambiguity high risk interval or S is less than the threshold S2, start the artificial review process.

8. A digital library index search system according to the digital library index search method of claim 7, characterized by: The system comprises: A query analysis module that receives a user input query text, performs language analysis on the query content, including dependency syntax analysis and key word extraction; A semantic shift evaluation module that uses a cross-language mapping model to calculate the shift degree of the core word in the multi-language semantic space and outputs a discretized drift risk level; The risk-driven compensation module judges whether to trigger the semantic anchor compensation operation according to the risk level, generates and adjusts the compensation coefficient, and establishes the semantic anchor network. The semantic matching scoring module calculates the semantic integrity score based on the anchor network and the semantic content of the query literature, and is used for judging the relevance and completeness of the search results. The decision control module makes a response operation according to the risk level and the score result, such as incremental training, manual review or default passing. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the digital library index retrieval method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the digital library index retrieval method of any one of claims 1-7.

Images