A knowledge retrieval method, device, and storage medium based on artificial intelligence

Abstract

This invention discloses an artificial intelligence-based knowledge retrieval method, apparatus, and storage medium to improve the precision and recall of knowledge retrieval and meet users' information retrieval needs. The method includes: receiving a user-submitted retrieval request, the retrieval request carrying a retrieval statement; extracting knowledge based on the retrieval statement to obtain entity information from the retrieval statement, the entity information including at least one of the following: entity, entity relationship, and entity attribute; constructing a concept based on the obtained entity information to obtain query terms; searching for target keywords matching the query terms from a pre-established document resource representation table, the document resource representation table containing data records of the mapping address information of the keywords and their respective original document resources; and extracting and presenting data records containing the target keywords from the document resource representation table.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology application technology, and in particular to a knowledge retrieval method, device and storage medium based on artificial intelligence. Background Technology

[0002] Currently, people primarily use search engines to retrieve information based on keywords. While this method can provide users with a large amount of information in a short time, the search results are often noisy, containing too much useless information. To obtain the information they truly need, users need to carefully analyze, identify, and select from the massive amount of feedback. Furthermore, users often don't know what keywords to enter when searching, only knowing what problem they've encountered, leading to inaccurate keyword input. In such cases, keyword-based search results fail to meet the user's actual needs. Moreover, searching based on user-submitted keywords has limitations, resulting in incomplete coverage of the returned search results.

[0003] Therefore, improving the precision and recall of search engine results to meet users' information retrieval needs has become one of the urgent technical problems to be solved in the current technology. Summary of the Invention

[0004] This invention provides a knowledge retrieval method, apparatus, and storage medium based on artificial intelligence, which can improve the precision and recall of knowledge retrieval and meet users' information retrieval needs.

[0005] Firstly, an artificial intelligence-based knowledge retrieval method is provided, including:

[0006] Receive a search request submitted by a user, wherein the search request carries a search statement;

[0007] Knowledge extraction is performed based on the search statement to obtain entity information in the search statement. The entity information includes at least one of the following: entity, entity relationship, and entity attribute.

[0008] Based on the obtained entity information, concepts are constructed to obtain query terms;

[0009] The target keywords that match the query terms are searched from a pre-established document resource representation table, which contains data records of the mapping address information of the keywords and their original document resources.

[0010] Extract and present data records containing the target keywords from the document resource representation table.

[0011] In one implementation, the document resource representation table also includes the membership degree corresponding to each keyword;

[0012] The document resource representation table was established according to the following process:

[0013] Knowledge extraction is performed on each original document resource to obtain document vocabulary;

[0014] For each document term, the membership degree of the document term is determined based on its first frequency of occurrence in the relevant document resources and its second frequency of being modified or connected by other document terms in the relevant document resources.

[0015] For each document resource, the keywords of that document resource are determined based on the membership degree of the document vocabulary contained therein.

[0016] For each keyword, establish the correspondence between the keyword and its corresponding membership degree and the mapping address of the original document resource to which it belongs to obtain the data record corresponding to the keyword;

[0017] The data records corresponding to the keywords of each original document resource are determined to form the document resource representation table.

[0018] In one implementation, for each document term, the membership degree of the document term is determined based on its first frequency of occurrence in the relevant document resource and its second frequency of being modified or connected by other document terms in the relevant document resource. Specifically, this includes:

[0019] For each document term, count the number of times the term appears in the relevant literature and the number of times the term is modified or linked by other terms in the relevant literature resources.

[0020] Based on the number of times the document's vocabulary appears in its respective literature and the total number of documents containing the vocabulary in its literature resources, determine the first frequency of the document's vocabulary in its respective literature resources;

[0021] Based on the number of times the document's vocabulary is modified or linked by other vocabulary in its respective document resources and the total number of document vocabulary contained in its respective document resources, determine the second frequency of the document's vocabulary being modified or linked by other document vocabulary in its respective document resources.

[0022] Based on the weight values ​​corresponding to the first and second frequencies, the membership degree of the document's vocabulary is determined according to the weighted result of the first and second frequencies.

[0023] In one implementation, before presenting data records containing the target keyword, the method further includes:

[0024] For each query term, a query term vector is formed by determining the membership degree corresponding to each query term based on the search statement; and

[0025] The membership degree of each target keyword that matches each query term is extracted from the document resource representation table to form a keyword vector, and the keyword vectors corresponding to each query term form a keyword matrix.

[0026] Determine the distance between the query term vector and each target keyword vector;

[0027] Based on the determined distances, the target keyword vectors contained in the keyword matrix are sorted; and

[0028] Extracting and presenting data records containing the target keywords, specifically including:

[0029] Based on the sorting results of each target keyword vector, extract and present the data records containing each target keyword corresponding to each target keyword vector.

[0030] In one implementation, for each target keyword corresponding to the target keyword vector, the data records of each target keyword are extracted and presented according to the following method:

[0031] For each target keyword vector, extract the data records containing each target keyword corresponding to that target keyword vector;

[0032] The extracted data records are sorted and presented according to the degree of membership of each target keyword.

[0033] In one implementation, the original document resources include video resources;

[0034] Before extracting the vocabulary from each original document resource, the process also includes:

[0035] Feature extraction is performed on video resources, and semantic description is obtained based on the extracted features to obtain video resource description information; and

[0036] Knowledge extraction is performed on each original document resource to obtain document vocabulary, specifically including:

[0037] For video resources, knowledge extraction is performed on the video resource description information corresponding to the video resources to obtain document vocabulary.

[0038] In one implementation, the original document resource includes audio resources;

[0039] Before extracting the vocabulary from each original document resource, the process also includes:

[0040] Speech recognition is performed on the speech resources to obtain text data; and

[0041] Knowledge extraction is performed on each original document resource to obtain document vocabulary, specifically including:

[0042] For audio resources, knowledge extraction is performed on the corresponding text data to obtain document vocabulary.

[0043] In one embodiment, the original document resources include original document resources in different languages; and

[0044] After obtaining the query terms by constructing concepts based on the acquired entity information, the following are also included:

[0045] Machine translation is used to convert the query terms into query terms in different languages; and

[0046] The search involves finding target keywords that match the query terms from a pre-established table of document resources, specifically including:

[0047] Search for target keywords in the corresponding languages ​​that match the query terms in different languages ​​from a pre-established table of document resources.

[0048] Secondly, an artificial intelligence-based knowledge retrieval device is provided, comprising:

[0049] The receiving unit is used to receive a search request submitted by a user, wherein the search request carries a search statement;

[0050] The acquisition unit is used to perform knowledge extraction based on the search statement to obtain entity information in the search statement, wherein the entity information includes at least one of the following: entity, entity relationship and entity attribute;

[0051] The concept building unit is used to build concepts based on the obtained entity information to obtain query terms;

[0052] The matching unit is used to search for target keywords that match the query terms from a pre-established document resource representation table, which contains data records of the mapping address information of the keywords and their original document resources.

[0053] The presentation unit is used to extract and present data records containing the target keywords from the document resource representation table.

[0054] In one implementation, the document resource representation table also includes the membership degree corresponding to each keyword;

[0055] The device further includes:

[0056] The mapping unit is used to extract knowledge from each original document resource to obtain document vocabulary; for each document vocabulary, the membership degree of the document vocabulary is determined based on its first frequency of occurrence in the document resource and its second frequency of being modified or connected by other document vocabulary in the document resource; for each document resource, the keywords of the document resource are determined based on the membership degree of the document vocabulary contained in the document resource; for each keyword, the correspondence between the keyword and its corresponding membership degree and the mapping address of its original document resource is established to obtain the data record corresponding to the keyword; the data records corresponding to the keywords of each original document resource are determined to form the document resource representation table.

[0057] In one implementation, the mapping unit is specifically configured to, for each document term, count the number of times the document term appears in its respective document and the number of times the document term is modified or connected by other terms in the document resource; determine a first frequency of the document term appearing in the document resource based on the number of times the document term appears in its respective document and the total number of document terms contained in the document resource; determine a second frequency of the document term being modified or connected by other document terms in the document resource based on the number of times the document term is modified or connected by other terms in the document resource and the total number of document terms contained in the document resource; and determine the membership degree of the document term based on the weight values ​​corresponding to the first frequency and the second frequency, according to the weighted result of the first frequency and the second frequency.

[0058] In one embodiment, the apparatus further includes a sorting unit, wherein:

[0059] The sorting unit is configured to, before the presentation unit presents data records containing the target keywords, establish a target keyword vector corresponding to each query term based on the membership degree of the query term and each target keyword matched in the document resource representation table for each query term. The target keyword vectors corresponding to each query term form a keyword matrix. In the target keyword vector, the vector component value corresponding to the target keyword that matches the query term is the membership degree of the keyword, and the vector component value corresponding to the keyword that does not match the query term is 0.

[0060] Determine the distance between the query term vector and the corresponding target keyword vector;

[0061] Based on the determined distance, the vectors of each target keyword contained in the keyword matrix are sorted.

[0062] The presentation unit is specifically used to extract and present data records containing each target keyword corresponding to each target keyword vector according to the sorting results of each target keyword vector.

[0063] In one embodiment, the presentation unit is further configured to extract data records containing each target keyword corresponding to each target keyword vector for each target keyword vector; and to sort and present the extracted data records according to the membership degree of each target keyword.

[0064] In one implementation, the original document resources include video resources;

[0065] The device further includes a feature extraction unit, wherein:

[0066] The feature extraction unit is used to extract features from video resources and perform semantic description based on the extracted features to obtain video resource description information.

[0067] The mapping unit is specifically used to extract document vocabulary from the video resource description information corresponding to the video resource.

[0068] In one implementation, the original document resource includes audio resources;

[0069] The device further includes a voice recognition unit, wherein:

[0070] The speech recognition unit is used to perform speech recognition on speech resources to obtain text data;

[0071] The mapping unit is specifically used to extract knowledge from the text data corresponding to the speech resource to obtain document vocabulary.

[0072] In one embodiment, the original document resources include original document resources in different languages; and

[0073] The device further includes a translation unit, wherein:

[0074] The translation unit is used to convert the query terms into query terms in different languages ​​using machine translation.

[0075] The matching unit is specifically used to search for target keywords in the corresponding language that match the query terms in different languages ​​from a pre-established document resource representation table.

[0076] Thirdly, a computing device is provided, comprising at least one processor and at least one memory, wherein the memory stores a computer program, and the processor is configured to read the computer program from the memory and execute any of the steps described in the artificial intelligence-based knowledge retrieval method provided in the first aspect.

[0077] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing computer-executable instructions for causing a computer to perform any of the steps described in the artificial intelligence-based knowledge retrieval method provided in the first aspect.

[0078] The knowledge retrieval method, apparatus, and storage medium based on artificial intelligence provided in this invention first extract keywords that can characterize the meaning of the original document resources to represent the original document resources, thereby obtaining a document resource representation table. On this basis, knowledge extraction is performed on the search statement submitted by the user to obtain entity information, which includes entities, entity relationships, and entity attributes. Further, concept establishment is performed on the entity information to obtain corresponding query terms. Since the query terms obtained after the above processing not only include the words contained in the search statement itself, but also include extended terms obtained by semantic understanding of these words based on the search statement, the target keywords obtained by matching the query terms in the document resource representation table can not only accurately express the meaning of the search statement, but also enrich the coverage of keywords, thereby improving the precision and recall of knowledge retrieval and meeting the user's information retrieval needs.

[0079] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0080] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0081] Figure 1 This is a schematic diagram illustrating the process of establishing a document resource representation table according to an embodiment of the present invention;

[0082] Figure 2 This is a schematic diagram illustrating the process of determining the membership degree of the document's vocabulary according to an embodiment of the present invention.

[0083] Figure 3This is a schematic diagram of the knowledge retrieval method based on artificial intelligence according to an embodiment of the present invention;

[0084] Figure 4 This is a schematic diagram of the process for sorting and displaying search results according to an embodiment of the present invention;

[0085] Figure 5 This is a schematic diagram of the structure of an artificial intelligence-based knowledge retrieval device according to an embodiment of the present invention. Detailed Implementation

[0086] To improve the recall and precision of knowledge retrieval and meet users' information retrieval needs, this invention provides a knowledge retrieval method, apparatus, and storage medium based on artificial intelligence.

[0087] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of the embodiments of the present invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein.

[0088] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention. Furthermore, the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0089] Example 1

[0090] The knowledge retrieval method based on artificial intelligence provided in this invention is applicable not only to the retrieval of document resources, but also to the retrieval of multimedia resources, such as video and audio resources.

[0091] In specific implementation, for document resources, in order to improve the precision and recall of knowledge retrieval, in this embodiment of the invention, document resources are categorized as follows: Figure 1 The process shown establishes a document resource representation table, including the following steps:

[0092] S11. Extract knowledge from the original document resources to obtain document vocabulary.

[0093] To reduce redundancy in the document and literature resource database and save on storage space, in practice, the document and literature resources are cleaned by data cleaning to remove incomplete, erroneous, and duplicate data before being stored in the original document and literature resource database.

[0094] For each original document resource contained in the original document resource library, the document vocabulary contained in that document resource is obtained through knowledge extraction.

[0095] S12. For each document term, determine the membership degree of the document term based on its first frequency of occurrence in the relevant document resources and its second frequency of being modified or connected by other document terms in the relevant document resources.

[0096] Among them, the membership degree of a document term refers to the ability of the document term to represent the document resource to which it belongs. The higher the membership degree, the better the document term can represent the document resource.

[0097] Based on the vocabulary of each document obtained in step S11, step S12 can be performed according to... Figure 2 The process shown determines the membership degree of the vocabulary in this document:

[0098] S121. For each document term, count the number of times the term appears in the relevant literature and the number of times the term is modified or linked by other terms in the relevant literature resources.

[0099] For each document term, for ease of description, let T be the number of times that term appears in the relevant literature. o Let T be the number of times a word in this document is modified or linked by other words in the corresponding document resource. c .

[0100] S122. Determine the first frequency of the document's vocabulary in the relevant literature based on the number of times the vocabulary appears in the relevant literature and the total number of vocabulary words in the relevant literature resources.

[0101] The frequency of the first occurrence of the vocabulary in this document within its associated literature resources is recorded as F. o In specific implementation, the first frequency F can be determined according to the following formula. o : Where T represents the total number of document terms contained in the literature to which the term belongs.

[0102] S123. Based on the number of times the document's vocabulary is modified or linked by other vocabulary in its respective document resource and the total number of document vocabulary contained in its respective document resource, determine the second frequency of the document's vocabulary being modified or linked by other document vocabulary in its respective document resource.

[0103] Let F be the second frequency of the vocabulary in this document being modified or linked by the vocabulary in other documents within the same document resource. c In practical implementation, the second frequency F can be determined according to the following formula. c :

[0104] S124. Based on the weight values ​​corresponding to the first frequency and the second frequency, determine the membership degree of the document's vocabulary according to the weighted result of the first frequency and the second frequency.

[0105] In practice, in order to improve the accuracy of membership degree calculation, in this embodiment of the invention, a manual correction value is also introduced in the membership degree calculation process, denoted as V∈[0,1], and its specific value can be obtained based on manual experience.

[0106] In this step, for any document term, the membership degree L can be determined using the following formula:

[0107]

[0108] Wherein, ω1 and ω2 are the weights of the first frequency of a document's vocabulary appearing in its respective document resource and the second weights of the frequency of a document's vocabulary being modified or connected by other words in its respective document resource, respectively.

[0109] S13. For each document resource, determine the keywords of that document resource based on the membership degree of the document vocabulary contained in that document resource.

[0110] In this step, for each document resource, several document terms that can represent the document resource can be selected as keywords in descending order of the degree of membership of the document terms contained in the document resource.

[0111] S14. For each keyword, establish the correspondence between the keyword and its corresponding membership degree and the mapping address of the original document resource to which it belongs to obtain the data record corresponding to the keyword.

[0112] In practice, the data record corresponding to each keyword can be represented as follows:

[0113] A={e i l i L(e i ), e i ∈U, i=1,2,...},e i l i L(e i ) is a composite element in the document resource representation table, e i These are the keywords extracted from the document's lexical domain U that represent the meaning of the entire document's documentary resources. i For the keyword e i The mapping address to the native document resource repository, L(e i (e) is the keyword i The degree of membership.

[0114] S15. Determine the data records corresponding to the keywords of each original document resource to form a document resource representation table.

[0115] In practice, the document resource representation table can also include fields such as descriptive information for the document resources. For document-type document resources, the descriptive information can be the document title.

[0116] In one embodiment, the structure of the document resource representation table is shown in Table 1:

[0117] Table 1

[0118] Keyl URL1 D1 Ll Key2 URL1 D1 L2 Key3 URL2 D2 L3 …… …… …… …… Keyn URLn Dn Ln

[0119] For video resources, in this embodiment of the invention, feature extraction is first performed on the video resources, and semantic description is then performed based on the extracted features to obtain video resource description information. It should be noted that, to reduce redundancy in the video resource description information and reduce storage resource overhead, in specific implementations, feature extraction can be performed on the video resources shot by shot, and semantic description of the video resource description information can be performed based on the extracted features. The resulting video resource description information can characterize the specific content of the video resource. After obtaining the video resource description information, through... Figure 2 The process shown establishes a document resource representation table corresponding to the video resource based on the video resource description information. For details, please refer to [link to documentation]. Figure 2 This will not be elaborated upon here.

[0120] The same method can be used to build an image resource representation table as for video resources. First, features are extracted from the image resources; then, semantic descriptions are performed based on the extracted features to obtain image resource description information; and finally, through... Figure 2 The process shown establishes a table representing the corresponding literature resources based on the image resource description information.

[0121] For audio resources, firstly, speech recognition is performed on the audio resources to obtain text data. Based on the text data corresponding to the audio resources, then... Figure 2 The process shown establishes a representation table of the corresponding literature resources for the speech resources. For details, please refer to [link / reference needed]. Figure 2 This will not be elaborated upon here.

[0122] In some embodiments, the document resource representation table may also include a resource type field to meet users' needs for retrieving different types of resources.

[0123] Through the above process, a document resource representation table containing document resources, video resources, and audio resources can be established. In order to improve the response speed of search results, in this embodiment of the invention, the document resource representation table can also be classified according to the type of each document resource. In this way, the search can be performed in the document resource representation classification table according to the search statement submitted by the user, thereby narrowing the search scope and improving the response speed of search results.

[0124] Based on the document resource representation table established above, this embodiment of the invention provides a knowledge retrieval method based on artificial intelligence, such as... Figure 3 As shown, it includes the following steps:

[0125] S31. Receive the search request submitted by the user.

[0126] In practice, when a user has a search need, they can submit a search request based on the problem they need to solve, including the search query. For ease of understanding, let's take the example of a user entering the search query "Why doesn't a microwave oven need to rotate anymore?" into a search engine.

[0127] S32. Based on the search statement, perform knowledge extraction to obtain entity information in the search statement.

[0128] The entity information includes at least one of the following: entity, entity relationship, and entity attribute.

[0129] Based on the search query submitted by the user in step S31, the entity information extracted using knowledge extraction methods is as follows: "microwave oven", "not needed", and "rotate". This step decomposes the search query into several words.

[0130] S33. Based on the obtained entity information, concepts are established to obtain query terms.

[0131] In this step, the concepts of each word obtained in step S32 are established, and synonyms or near-synonyms of each word are obtained. For example, "rotate" can be expanded to "turn," "flip," "twist," etc. The original words obtained in step S32 and the expanded words obtained in step S33 are used as query words. Continuing the example above, the query words can be represented as "microwave oven," "not needed," "rotate," "turn," "flip," and "spin."

[0132] S34. Search for target keywords that match the query terms from a pre-established table of document resources.

[0133] In practice, this step involves using the query terms obtained in step S33 to search for matching target keywords in Table 1 or the document resource representation classification table. In this example, the target keywords matched by the query terms in Table 1 are "Key1", "Key3", etc.

[0134] S35. Extract and present data records containing target keywords from the document resource representation table.

[0135] In this step, data records containing "Key1", "Key3", etc., are extracted from Table 1 and presented to the user, as shown in Table 2:

[0136] Table 2

[0137] Key1 URL1 D1 L1 Key3 URL2 D2 L3 …… …… …… ……

[0138] Example 1 provides a concept-based abstract retrieval method, which differs from existing keyword-based retrieval methods. This method can retrieve literature that can solve user problems but does not contain keywords in the search query. In addition, the retrieval method provided by this embodiment can also provide multimedia literature resources, while existing retrieval methods can only retrieve structured and semi-structured text data and cannot retrieve multimedia unstructured data such as images, videos, and audio. This improves the precision and recall of knowledge retrieval.

[0139] Example 2

[0140] In specific implementation, when presenting the search results obtained in Example 1, in order to display search results that are closer to the user's search intent first and reduce the time the user spends searching for literature resources that can solve their problem, Example 2, based on the search results obtained in Example 1, sorts and displays the retrieved literature resources according to the distance between each literature resource and the user's search intent. In this embodiment of the invention, the user's search intent can be represented by the membership degree of the query terms obtained in step S33, and the main content of each literature resource can be represented by the membership degree of the keywords. Therefore, in Example 2, the search results can be sorted by calculating the distance between the membership degree of the query terms and the membership degree between the keywords.

[0141] Step S124 above describes the membership degree of keywords in document resources. The calculation method for the membership degree of query terms is based on the same principle as the calculation method for the membership degree of keywords, such as... Figure 4 As shown, for each query term, the membership degree of that query term can be calculated using the following process:

[0142] L=ω3*F a +ω4*F b

[0143] Among them, F a F indicates the third frequency of the query term in the search query. b This indicates the fourth frequency of the query term being modified or connected by other terms in the search query. ω3 and ω4 represent F... a and F b The weighting can be set according to actual needs or based on empirical values ​​during implementation. It should be noted that in this embodiment of the invention, the query terms include extended terms of the original terms extracted from the search statement. Therefore, when calculating the third frequency, the number of times the query terms appear can be counted as the number of times the original terms and their extended terms extracted from the search statement appear. Similarly, when calculating the fourth frequency, the number of times the query terms are modified or connected by other terms in the search statement can be counted as the number of times the original terms and their extended terms included in the search statement modify or connect them.

[0144] Using the above method, the membership degrees of keywords and query terms can be obtained. Based on the determined membership degrees of query terms and keywords, in this embodiment of the invention, it can be done according to... Figure 4 The process shown sorts and displays the search results:

[0145] S41. For each query term, the membership degree corresponding to each query term is determined based on the search statement to form a query term vector.

[0146] Specifically, the query word vector can be represented as q = (L(q1), L(q2), L(q3), ..., L(q...). n )) T q i Let i = 1, 2, ..., n be the search keywords, and l(q) i ) is q i The degree of membership.

[0147] S42. For each query term, establish a target keyword vector corresponding to the query term based on the membership degree between the query term and each target keyword matched in the document resource representation table. The target keyword vectors corresponding to each query term form a keyword matrix.

[0148] In the target keyword vector, the vector component value corresponding to the target keyword that matches the query keyword is the membership degree of the keyword, and the vector component value corresponding to the keyword that does not match the query keyword is 0.

[0149] For example, for query term q1, its corresponding target keyword vector can be represented as [L(r 11 ), 0, L(r) 13), 0, 0, 0......L(r) m1 ]], where m represents the number of keywords contained in the document resource representation table. In specific implementation, the query term q is used. i Match the target keywords in the document resource representation table, and in the query term q i With the keyword r m In the case of a match, the value of the corresponding vector component in the keyword vector is the membership degree of the matched keyword. In the query term q... i With the keyword r m In the case of a mismatch, the corresponding vector component in the keyword vector has a value of 0. Accordingly, the keyword matrix can be represented as... r ij To search for the word q i The corresponding j-th keyword, L(r) ij ) is the keyword r ij The membership degree is given by , where n is the number of query keywords. A column in R represents a keyword vector consisting of the membership degrees of keywords matching a given query term.

[0150] In one embodiment, r ij The following principles can be used as a guide for selection:

[0151] R = {r ij , L(q i )≤L(r ij ), r ij ∈L and r ij =q i}, q i ∈q, where L(q) i L(r) represents the membership degree of the query term. ij ) represents the keyword membership degree, L represents the document resource representation table, and R represents the query matching results.

[0152] For example, keyword vectors can be represented as:

[0153]

[0154] S43. Determine the distance between the query term vector and the vector of each target keyword.

[0155] In one embodiment, the distance between the query word vector and the target keyword vector can be represented using cosine distance. Let r j Let R be the j-th row, where j = 1, 2, ..., n, representing the keyword vector corresponding to the membership degree of the keyword matching the j-th query term. In this embodiment of the invention, the cosine distance between the query term vector and the target keyword vector can be calculated using the following formula:

[0156]

[0157] Where, q k It is the k-th component of the query word vector q, r kj It is the keyword vector r k The j-th component, k∈[1,n], j∈[1,m], θ j It is the angle between vectors.

[0158] S44. Based on the determined distance, sort the vectors of each target keyword in the keyword matrix.

[0159] In this step, that is, according to cos(θ) j Arrange r in descending order k .

[0160] For the keyword vector r k Each element contained herein, in this embodiment of the invention, can be arranged according to L(r) ij Arrange them in descending order to form a vector r', and the new vector r' is the sorted result.

[0161] It should be noted that for the keyword vector r k The sorting of each element according to its membership degree is only one implementation method of this invention. In specific implementations, the elements can also be sorted according to the access volume, publication time, etc. of the corresponding literature resources. This invention does not limit this method.

[0162] Based on the sorting results obtained above, in specific implementation, step S35 can extract and present data records containing each target keyword corresponding to each target keyword vector according to the sorting results of each target keyword vector.

[0163] Taking the keywords matched by query term 1 as Key1, Key2, Key3, Key4, Key5, and the keywords matched by query term 2 as Key6, Key7, Key8, Key9, Key10 as an example, in specific implementation, the cosine distances calculated according to step S43 are cos(θ1) and cos(θ2), and cos(θ1) is less than cos(θ2). The data records extracted according to the sorting results obtained in step S44 are shown in Table 3:

[0164] Table 3

[0165] Key6 URL5 D4 L6 Key7 URL4 D4 L7 Key8 URL6 D3 L8 Key9 URL9 D6 L9 Key10 URL9 D6 L10 Key| URL1 D1 L1 Key2 URL1 D1 L2 Key3 URL2 D2 L3 Key4 URL3 D5 L4 Key5 URL7 D7 L5

[0166] It should be noted that if the extracted data records contain the same URL address, the duplicate records will be deleted, and only one record will be kept.

[0167] Furthermore, if the membership degree ranking of the keywords matching query term 2 is L8 < L6 < L7 < L9 < L10, and the membership degree ranking of the keywords matching query term 1 is L5 < L1 < L4 < L2 < L3, then the extracted data records are shown in Table 4:

[0168] Table 4

[0169] Key10 URL9 D6 L10 Key9 URL9 D6 L9 Key7 URL4 D4 L7 Key6 URL5 D4 L6 Key8 URL6 D3 L8 Key3 URL2 D2 L3 Key2 URL1 D1 L2 Key4 URL3 D5 L4 Key1 URL1 D1 L1 Key5 URL7 D7 L5

[0170] Based on the presented search results, users can view the original literature resources via URL.

[0171] According to the AI-based retrieval method provided in Embodiment 2, the retrieval results obtained in the embodiment can be sorted and displayed according to the user's true intent, reducing the time for the user to locate the required literature resources and improving retrieval efficiency.

[0172] Example 3

[0173] To meet users' needs for multilingual document resource retrieval, Example 3, based on Examples 1 and 2, allows machine translation to be used to translate the query terms into different languages ​​after obtaining them. Then, target keywords matching the query terms in different languages ​​can be found from a pre-established document resource representation table. This achieves cross-language knowledge retrieval, meets users' different language retrieval needs, and further improves the recall rate of knowledge retrieval.

[0174] Based on the same technical concept, this application also provides an artificial intelligence-based knowledge retrieval device. Since the principle of the above device in solving the problem is similar to that of the artificial intelligence-based knowledge retrieval method, the implementation of the above device can refer to the implementation of the method, and the repeated parts will not be described again.

[0175] like Figure 5 As shown, it is a schematic diagram of the structure of an artificial intelligence-based knowledge retrieval device provided in an embodiment of the present invention, including:

[0176] The receiving unit 51 is used to receive a search request submitted by a user, wherein the search request carries a search statement;

[0177] The acquisition unit 52 is used to perform knowledge extraction based on the search statement to obtain entity information in the search statement. The entity information includes at least one of the following: entity, entity relationship, and entity attribute.

[0178] Concept building unit 53 is used to build concepts based on the obtained entity information to obtain query terms;

[0179] The matching unit 54 is used to search for target keywords that match the query terms from a pre-established document resource representation table, wherein the document resource representation table contains data records of the mapping address information of the keywords and their original document resources.

[0180] Presentation unit 55 is used to extract and present data records containing the target keywords from the document resource representation table.

[0181] In one implementation, the document resource representation table also includes the membership degree corresponding to each keyword;

[0182] The device further includes:

[0183] The mapping unit is used to extract knowledge from each original document resource to obtain document vocabulary; for each document vocabulary, the membership degree of the document vocabulary is determined based on its first frequency of occurrence in the document resource and its second frequency of being modified or connected by other document vocabulary in the document resource; for each document resource, the keywords of the document resource are determined based on the membership degree of the document vocabulary contained in the document resource; for each keyword, the correspondence between the keyword and its corresponding membership degree and the mapping address of its original document resource is established to obtain the data record corresponding to the keyword; the data records corresponding to the keywords of each original document resource are determined to form the document resource representation table.

[0184] In one implementation, the mapping unit is specifically configured to, for each document term, count the number of times the document term appears in its respective document and the number of times the document term is modified or connected by other terms in the document resource; determine a first frequency of the document term appearing in the document resource based on the number of times the document term appears in its respective document and the total number of document terms contained in the document resource; determine a second frequency of the document term being modified or connected by other document terms in the document resource based on the number of times the document term is modified or connected by other terms in the document resource and the total number of document terms contained in the document resource; and determine the membership degree of the document term based on the weight values ​​corresponding to the first frequency and the second frequency, according to the weighted result of the first frequency and the second frequency.

[0185] In one embodiment, the apparatus further includes a sorting unit, wherein:

[0186] The sorting unit is configured to, before the presentation unit presents data records containing the target keywords, determine the membership degree of each query term based on the search statement to form a query term vector for each query term; extract the membership degree of each target keyword matching each query term from the document resource representation table to form a keyword vector, and form a keyword matrix from the keyword vectors corresponding to each query term; determine the distance between each query term vector and each target keyword vector; and sort the target keyword vectors contained in the keyword matrix according to the determined distance.

[0187] The presentation unit is specifically used to extract and present data records containing each target keyword corresponding to each target keyword vector according to the sorting results of each target keyword vector.

[0188] In one embodiment, the presentation unit is further configured to extract data records containing each target keyword corresponding to each target keyword vector for each target keyword vector; and to sort and present the extracted data records according to the membership degree of each target keyword.

[0189] In one implementation, the original document resources include video resources;

[0190] The device further includes a feature extraction unit, wherein:

[0191] The feature extraction unit is used to extract features from video resources and perform semantic description based on the extracted features to obtain video resource description information.

[0192] The mapping unit is specifically used to extract document vocabulary from the video resource description information corresponding to the video resource.

[0193] In one implementation, the original document resource includes audio resources;

[0194] The device further includes a voice recognition unit, wherein:

[0195] The speech recognition unit is used to perform speech recognition on speech resources to obtain text data;

[0196] The mapping unit is specifically used to extract knowledge from the text data corresponding to the speech resource to obtain document vocabulary.

[0197] In one embodiment, the original document resources include original document resources in different languages; and

[0198] The device further includes a translation unit, wherein:

[0199] The translation unit is used to convert the query terms into query terms in different languages ​​using machine translation.

[0200] The matching unit is specifically used to search for target keywords in the corresponding language that match the query terms in different languages ​​from a pre-established document resource representation table.

[0201] For ease of description, the above sections are divided into modules (or units) according to their functional modules and described separately. Of course, in implementing this invention, the functions of each module (or unit) can be implemented in one or more software or hardware components.

[0202] After introducing the artificial intelligence-based knowledge retrieval method and apparatus according to exemplary embodiments of the present invention, a computing device according to another exemplary embodiment of the present invention will be introduced next.

[0203] Those skilled in the art will understand that various aspects of the present invention can be implemented as systems, methods, or program products. Therefore, various aspects of the present invention can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system." In some possible embodiments, the computing device according to the present invention may include at least one processor and at least one memory. The memory stores program code that, when executed by the processor, causes the processor to perform the steps in the artificial intelligence-based knowledge retrieval method according to various exemplary embodiments of the present invention described above. For example, the processor may perform, as... Figure 3 The steps shown are: S31, receiving a search request submitted by a user; S32, extracting knowledge based on the search statement to obtain entity information in the search statement; S33, establishing concepts based on the obtained entity information to obtain query terms; S34, searching for target keywords that match the query terms from a pre-established document resource representation table; and S35, extracting and presenting data records containing target keywords from the document resource representation table.

[0204] In some possible implementations, various aspects of the AI-based knowledge retrieval method provided by the present invention can also be implemented as a program product, comprising program code. When the program product is run on a computer device, the program code causes the computer device to execute the steps in the AI-based knowledge retrieval method according to various exemplary embodiments of the present invention described above. For example, the computer device can execute, for example, Figure 3The steps shown are: S31, receiving a search request submitted by a user; S32, extracting knowledge based on the search statement to obtain entity information in the search statement; S33, establishing concepts based on the obtained entity information to obtain query terms; S34, searching for target keywords that match the query terms from a pre-established document resource representation table; and S35, extracting and presenting data records containing target keywords from the document resource representation table.

[0205] The program product can take the form of any combination of one or more readable media. A readable medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. The program product of embodiments of the present invention for XX can take the form of a portable compact disk read-only memory (CD-ROM) and include program code, and can run on a computing device. However, the program product of the present invention is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0206] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0207] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that specifies a function in one or more boxes.

[0208] These computer program instructions may also be stored in a computer-readable storage medium that directs a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0209] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps that specify the function are in one or more boxes.

[0210] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0211] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A knowledge retrieval method based on artificial intelligence, characterized in that, include: Receive a search request submitted by a user, wherein the search request carries a search statement; Knowledge extraction is performed based on the search statement to obtain entity information in the search statement. The entity information includes at least one of the following: entity, entity relationship, and entity attribute. Based on the obtained entity information, concepts are constructed to obtain query terms; The system searches for target keywords that match the query terms from a pre-established document resource representation table. The document resource representation table contains data records of the mapping addresses of the keywords and their respective original document resources. It also contains the membership degree of each keyword. The document resource representation table is established according to the following process: knowledge extraction is performed on each original document resource to obtain document terms; for each document term, the membership degree is determined based on its first frequency of occurrence in its respective document resource and its second frequency of being modified or connected by other document terms in its respective document resource. Specifically, this includes: for each document term, counting the number of times it appears in its respective document resource and the number of times it is modified or connected by other terms in its respective document resource. Based on the frequency of the vocabulary in this document in the relevant literature The total number of documents and their associated literature resources. The first frequency of a document's vocabulary within its corresponding literature resources is determined using the following formula. : ; Based on the number of times the vocabulary in this document is modified or linked by other vocabulary in the relevant literature resources. The total number of documents and their associated literature resources. The second frequency of a document's vocabulary being modified or linked by other document vocabulary within its own literature resource is determined using the following formula. : ; Based on the weight values ​​corresponding to the first and second frequencies, and according to the weighted result of the first and second frequencies, the membership degree of the document's vocabulary is determined according to the following formula: ； Wherein, ω1 and ω2 are the weights of the first frequency of a document word appearing in its respective document resource and the second weights of the frequency of a document word being modified or connected by other words in its respective document resource, respectively. ∈[0,1], is a manually corrected value, the specific value of which can be obtained based on human experience; For each document resource, the keywords of the document resource are determined based on the membership degree of the document vocabulary contained in the document resource; for each keyword, the correspondence between the keyword and its corresponding membership degree and the mapping address of the original document resource to which it belongs is established to obtain the data record corresponding to the keyword; the data records corresponding to the keywords of each original document resource are determined to form the document resource representation table. For each query term, a query term vector is formed by determining the membership degree corresponding to each query term based on the search statement; and For each query term, a target keyword vector corresponding to the query term is established based on the membership degree of the query term and each target keyword that matches the document resource representation table. The target keyword vectors corresponding to each query term form a keyword matrix. In the target keyword vector, the vector component value corresponding to the target keyword that matches the query term is the membership degree of the keyword, and the vector component value corresponding to the keyword that does not match the query term is 0. Determine the distance between the query term vector and the corresponding target keyword vector; Based on the determined distance, the vectors of each target keyword contained in the keyword matrix are sorted. Extracting and presenting data records containing the target keywords from the document resource representation table, specifically including: Based on the sorting results of each target keyword vector, extract and present the data records containing each target keyword corresponding to each target keyword vector.

2. The method according to claim 1, characterized in that, For each target keyword corresponding to the target keyword vector, the data records of each target keyword are extracted and presented according to the following method: For each target keyword vector, extract the data records containing each target keyword corresponding to that target keyword vector; The extracted data records are sorted and presented according to the degree of membership of each target keyword.

3. The method according to claim 1, characterized in that, The original document resources include video resources; Before extracting the vocabulary from each original document resource, the process also includes: Feature extraction is performed on video resources, and semantic description is obtained based on the extracted features to obtain video resource description information; and Knowledge extraction is performed on each original document resource to obtain document vocabulary, specifically including: For video resources, knowledge extraction is performed on the video resource description information corresponding to the video resources to obtain document vocabulary.

4. The method according to claim 1, characterized in that, The original document resources include audio resources; Before extracting the vocabulary from each original document resource, the process also includes: Speech recognition is performed on the speech resources to obtain text data; and Knowledge extraction is performed on each original document resource to obtain document vocabulary, specifically including: For audio resources, knowledge extraction is performed on the corresponding text data to obtain document vocabulary.

5. The method according to claim 1, characterized in that, The original document resources include original document resources in different languages; as well as After obtaining the query terms by constructing concepts based on the acquired entity information, the following are also included: Machine translation is used to convert the query terms into query terms in different languages; and The search involves finding target keywords that match the query terms from a pre-established table of document resources, specifically including: Search for target keywords in the corresponding languages ​​that match the query terms in different languages ​​from a pre-established table of document resources.

6. A knowledge retrieval device based on artificial intelligence, characterized in that, The apparatus is used to implement the method according to claims 1-5, and the apparatus comprises: The receiving unit is used to receive a search request submitted by a user, wherein the search request carries a search statement; The acquisition unit is used to perform knowledge extraction based on the search statement to obtain entity information in the search statement, wherein the entity information includes at least one of the following: entity, entity relationship and entity attribute; The concept building unit is used to build concepts based on the obtained entity information to obtain query terms; A matching unit is used to search for target keywords that match the query terms from a pre-established document resource representation table. The document resource representation table contains data records of the mapping address information of the keywords and their respective original document resources. The document resource representation table also contains the membership degree corresponding to each keyword. The document resource representation table is established according to the following process: knowledge extraction is performed on each original document resource to obtain document terms; for each document term, the membership degree of the document term is determined based on the first frequency of the document term appearing in its respective document resource and the second frequency of the document term being modified or connected by other document terms in its respective document resource. Specifically, this includes: for each document term, counting the number of times the document term appears in its respective document resource and the number of times the document term is modified or connected by other terms in its respective document resource. Based on the frequency of the vocabulary in this document in the relevant literature The total number of documents and their associated literature resources. The first frequency of a document's vocabulary within its corresponding literature resources is determined using the following formula. : ; Based on the number of times the vocabulary in this document is modified or linked by other vocabulary in the relevant literature resources. The total number of documents and their associated literature resources. The second frequency of a document's vocabulary being modified or linked by other document vocabulary within its own literature resource is determined using the following formula. : ; Based on the weight values ​​corresponding to the first and second frequencies, and according to the weighted result of the first and second frequencies, the membership degree of the document's vocabulary is determined according to the following formula: ； Wherein, ω1 and ω2 are the weights of the first frequency of a document word appearing in its respective document resource and the second weights of the frequency of a document word being modified or connected by other words in its respective document resource, respectively. ∈[0,1], is a manually corrected value, the specific value of which can be obtained based on human experience; For each document resource, the keywords of the document resource are determined based on the membership degree of the document vocabulary contained in the document resource; for each keyword, the correspondence between the keyword and its corresponding membership degree and the mapping address of the original document resource to which it belongs is established to obtain the data record corresponding to the keyword; the data records corresponding to the keywords of each original document resource are determined to form the document resource representation table. The presentation unit is used to extract and present data records containing the target keywords from the document resource representation table.

7. A computing device, characterized in that, It includes at least one processor and at least one memory, wherein the memory stores a computer program, and the processor is configured to read the computer program from the memory and execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the method described in any one of claims 1 to 5.

Images