Text processing method, device, equipment, storage medium and program product

Abstract

The embodiment of the present application provides a text processing method, device, equipment, storage medium and program product. The method includes: obtaining a query text input by a user and obtaining a cluster summary tree; performing multi-level search processing on the cluster summary tree according to the query text to obtain at least one target node corresponding to the query text; generating user reference information according to the query text and the business knowledge text of each target node; performing language processing on the reference information to obtain the user's reply information. The method is used to achieve the effect of improving the accuracy of text recognition results.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a text processing method, device, equipment, storage medium and program product. Background Art

[0002] In the process of handling error business and other business, it is necessary to respond to business inquiries raised by users.

[0003] In the prior art, a large language model is used to process query texts such as business query requirements input by users, and generate response information to respond to the business requirements raised by the users.

[0004] However, in the above method, the large language model has the problem of untimely knowledge update, which affects the accuracy of the text processing results based on the large language model. Summary of the invention

[0005] The embodiments of the present application provide a text processing method, apparatus, device, storage medium and program product to improve the accuracy of text recognition results.

[0006] In a first aspect, an embodiment of the present application provides a text processing method, comprising:

[0007] Obtaining a query text input by a user and obtaining a cluster summary tree; wherein the query text represents the user's business knowledge query demand; wherein the cluster summary tree includes at least one node; and the node stores the business knowledge text;

[0008] According to the query text, the cluster summary tree is subjected to multi-level retrieval processing to obtain at least one target node corresponding to the query text; wherein the target node is a node in the cluster summary tree associated with the query text;

[0009] Generate reference information of the user according to the query text and the business knowledge text of each target node; wherein the reference information represents the business query demand of the user;

[0010] The reference information is language processed to obtain the user's reply information; wherein the reply information includes the business knowledge required by the user.

[0011] In a possible implementation, obtaining a cluster summary tree includes:

[0012] Acquire a business document, and segment the business document to obtain at least one slice corresponding to the business document; wherein the slice represents a business knowledge text;

[0013] The cluster summary tree is generated according to each of the slices.

[0014] In a possible implementation, generating the cluster summary tree according to each of the slices includes:

[0015] Generate a leaf node corresponding to the slice; wherein the leaf node stores the business knowledge text represented by the slice;

[0016] Performing clustering processing on each of the slices to obtain at least one first cluster; wherein the slices in the first cluster have similar semantics;

[0017] Processing each of the first clusters to obtain a parent node set; wherein the parent node set includes at least one parent node; the parent node stores the business knowledge text represented by the first cluster;

[0018] Wherein, each leaf node and each parent node constitute the cluster summary tree.

[0019] In a possible implementation, clustering is performed on each of the slices to obtain at least one first cluster, including:

[0020] Determine the vector distance between every two slice vectors in the preset vector space among the slice vectors corresponding to each of the slices;

[0021] According to each of the vector distances, vector clustering processing is performed on each of the slice vectors to obtain each of the first clusters.

[0022] In a possible implementation, each of the first clusters is processed to obtain a parent node set, including:

[0023] Performing language processing on each of the first clusters to obtain a summary set corresponding to each of the first clusters; wherein the summary set includes a summary corresponding to each of the first clusters; and the summary is a text summary of each of the business knowledge texts;

[0024] The parent node set is generated according to each summary in the summary set.

[0025] In a possible implementation, generating the parent node set according to each digest in the digest set includes:

[0026] Generate an i-th layer node set according to the i-th summary set; wherein the node set includes the parent node corresponding to each summary in the summary set;

[0027] Performing vector clustering processing on the abstract vectors corresponding to the abstracts in the i-th abstract set to obtain the i-th cluster set; wherein the i-th cluster set includes at least one second cluster; and the abstracts in the second cluster have similar semantics;

[0028] Performing language processing on each second cluster in the i-th cluster set to obtain an i+1-th summary set; wherein the i+1-th summary set includes a summary corresponding to each second cluster;

[0029] According to the i+1th summary set, an i+1th layer node set is generated; wherein each of the node sets constitutes the parent node set.

[0030] In a possible implementation, the cluster summary tree includes at least one leaf node and at least one parent node; according to the query text, a multi-level search process is performed on the cluster summary tree to obtain at least one target node corresponding to the query text, including:

[0031] Determine at least one starting node corresponding to the query text from each leaf node in the cluster summary tree;

[0032] According to each of the starting nodes, a recursive search process is performed on each of the parent nodes in the cluster summary tree to obtain each of the target nodes.

[0033] In a possible implementation, determining at least one starting node corresponding to the query text from each leaf node in the cluster summary tree includes:

[0034] According to the query text, keyword retrieval processing is performed on each leaf node to obtain a first candidate document corresponding to the query text; wherein the first candidate document includes at least one business knowledge text in a leaf node that matches the query text;

[0035] According to the query text, semantic retrieval processing is performed on each leaf node to obtain a second candidate document corresponding to the query text; wherein the second candidate document includes at least one business knowledge text in a leaf node that is semantically similar to the query text;

[0036] According to the first candidate document and the second candidate document, each of the starting nodes is determined from each of the leaf nodes.

[0037] In a possible implementation, performing semantic retrieval processing on each leaf node according to the query text to obtain a second candidate document corresponding to the query text includes:

[0038] Determine the cosine similarity between the query vector corresponding to the query text and the node vector in each leaf node; wherein the node vector is the vector corresponding to the business knowledge text in the leaf node;

[0039] The second candidate document is determined according to each of the cosine similarities.

[0040] In a possible implementation, determining each of the starting nodes from each of the leaf nodes according to the first candidate document and the second candidate document includes:

[0041] Determine first relevant information between the first candidate document and the query text; and determine second relevant information between the second candidate document and the query text; wherein the first relevant information represents the correlation between each business knowledge text in the first candidate document and the query text; and the second relevant information represents the correlation between each business knowledge text in the second candidate document and the query text;

[0042] Determine the best candidate document according to the first relevant information and the second relevant information; wherein the best candidate document includes at least one business knowledge text associated with the query text;

[0043] The leaf node corresponding to each business knowledge text in the best candidate document is determined from each leaf node as the starting node.

[0044] In a possible implementation, the method further includes:

[0045] If it is determined that there is no target node corresponding to the query text in the cluster summary tree, preset default response information is generated.

[0046] In a second aspect, an embodiment of the present application provides a text processing device, comprising: an acquisition module, used to acquire a query text input by a user, and acquire a cluster summary tree; wherein the query text represents the user's business knowledge query demand; wherein the cluster summary tree includes at least one node; and the node stores business knowledge text;

[0047] A retrieval module, configured to perform a multi-level retrieval process on the cluster summary tree according to the query text to obtain at least one target node corresponding to the query text; wherein the target node is a node in the cluster summary tree associated with the query text;

[0048] A generating module, used to generate reference information of the user according to the query text and the business knowledge text of each target node; wherein the reference information represents the business query demand of the user;

[0049] The reply module is used to perform language processing on the reference information to obtain the user's reply information; wherein the reply information includes the business knowledge required by the user.

[0050] In a possible implementation, the acquisition module is specifically used to: acquire a business document, and segment the business document to obtain at least one slice corresponding to the business document; wherein the slice represents the business knowledge text; and generate the cluster summary tree based on each of the slices.

[0051] In a possible implementation, the acquisition module is specifically used to: generate a leaf node corresponding to the slice; wherein the leaf node stores the business knowledge text represented by the slice; cluster each of the slices to obtain at least one first cluster cluster; wherein the slices in the first cluster cluster have similar semantics; process each of the first cluster clusters to obtain a parent node set; wherein the parent node set includes at least one parent node; the parent node stores the business knowledge text represented by the first cluster cluster; wherein each of the leaf nodes and each of the parent nodes constitute the cluster summary tree.

[0052] In a possible implementation, the acquisition module is specifically used to: determine the vector distance between every two slice vectors in the preset vector space among the slice vectors corresponding to each of the slices; and perform vector clustering processing on each of the slice vectors according to each of the vector distances to obtain each of the first cluster clusters.

[0053] In a possible implementation, the acquisition module is further specifically used to: perform language processing on each of the first clusters to obtain a summary set corresponding to each of the first clusters; wherein the summary set includes a summary corresponding to each of the first clusters; the summary is a text summary of each business knowledge text; and generate the parent node set based on each summary in the summary set.

[0054] In a possible implementation, the acquisition module is further specifically used to: generate an i-th layer node set according to the i-th summary set; wherein the node set includes the parent node corresponding to each summary in the summary set; perform vector clustering processing on the summary vectors corresponding to each summary in the i-th summary set to obtain the i-th cluster cluster set; wherein the i-th cluster cluster set includes at least one second cluster cluster; the summaries in the second cluster clusters are semantically similar; perform language processing on each second cluster cluster in the i-th cluster cluster set to obtain the i+1-th summary set; wherein the i+1-th summary set includes the summaries corresponding to each second cluster cluster; generate an i+1-th layer node set according to the i+1-th summary set; wherein each of the node sets constitutes the parent node set.

[0055] In a possible implementation, the cluster summary tree includes at least one leaf node and at least one parent node; the retrieval module is specifically used to: determine at least one starting node corresponding to the query text from each leaf node in the cluster summary tree; and perform recursive retrieval processing on each parent node in the cluster summary tree according to each starting node to obtain each target node.

[0056] In a possible implementation, the retrieval module is specifically used to: perform keyword retrieval processing on each of the leaf nodes according to the query text to obtain a first candidate document corresponding to the query text; wherein the first candidate document includes at least one business knowledge text in a leaf node that matches the query text; perform semantic retrieval processing on each of the leaf nodes according to the query text to obtain a second candidate document corresponding to the query text; wherein the second candidate document includes at least one business knowledge text in a leaf node that is semantically similar to the query text; and determine each of the starting nodes from each of the leaf nodes according to the first candidate document and the second candidate document.

[0057] In a possible implementation, the retrieval module is specifically used to: determine the cosine similarity between the query vector corresponding to the query text and the node vector in each leaf node; wherein the node vector is the vector corresponding to the business knowledge text in the leaf node; and determine the second candidate document based on each cosine similarity.

[0058] In a possible implementation, the retrieval module is further specifically used to: determine the first relevant information between the first candidate document and the query text; and determine the second relevant information between the second candidate document and the query text; wherein the first relevant information represents the correlation between each business knowledge text in the first candidate document and the query text; the second relevant information represents the correlation between each business knowledge text in the second candidate document and the query text; determine the best candidate document based on the first relevant information and the second relevant information; wherein the best candidate document includes at least one business knowledge text associated with the query text; and determine the leaf node corresponding to each business knowledge text in the best candidate document from each of the leaf nodes as the starting node.

[0059] In a possible implementation manner, the device is further configured to: generate preset default response information if it is determined that there is no target node corresponding to the query text in the cluster summary tree.

[0060] In a third aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

[0061] The memory stores computer-executable instructions;

[0062] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0063] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementations of the first aspect.

[0064] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0065] The text processing method, device, equipment, storage medium and program product provided in the embodiments of the present application obtain the query text input by the user and obtain the cluster summary tree; perform multi-level search processing on the cluster summary tree according to the query text to obtain at least one target node corresponding to the query text; generate the user's reference information according to the query text and the business knowledge text of each target node; perform language processing on the reference information to obtain the user's reply information. By performing multi-level search processing on the cluster summary tree according to the query text input by the user, the nodes associated with the query text in the cluster summary tree are obtained, and the user's reference information is generated according to the query text and the business knowledge text in the nodes associated with the query text, so as to perform natural language processing and obtain the business knowledge required by the user; thereby, the accuracy of the text recognition result can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0067] Figure 1 A schematic diagram of an application scenario provided for this application;

[0068] Figure 2 A flowchart of a text processing method provided in an embodiment of the present application;

[0069] Figure 3 A flowchart of another text processing method provided in an embodiment of the present application;

[0070] Figure 4 A data processing and storage flow chart provided for an embodiment of the present application;

[0071] Figure 5 A schematic diagram of generating a cluster summary tree provided in an embodiment of the present application;

[0072] Figure 6 A question-answering flow chart based on a cluster summary tree provided in an embodiment of the present application;

[0073] Figure 7 A schematic diagram of a bottom-up progressive multi-level retrieval method provided in an embodiment of the present application;

[0074] Figure 8 A structural diagram of a text processing device provided in an embodiment of the present application;

[0075] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0076] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0077] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0078] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0079] In addition, this application involves conducting big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and using artificial intelligence technology to make automated decisions, and making technical solutions that have a significant impact on personal rights and interests based on the results of automated decisions. The application provides users with corresponding operation entrances for them to choose to agree or reject the results of automated decisions; if the user chooses to reject, the expert decision-making process will be entered.

[0080] It should be noted that the present application can be used in the field of artificial intelligence, and can also be used in any field other than artificial intelligence, and the application field of the present application is not limited.

[0081] Figure 1 A schematic diagram of an application scenario provided for this application, such as Figure 1 As shown, the specific application scenario of the present application includes an electronic device 101 and a user 102. In the process of processing error services and other services based on the electronic device 101, it is necessary to respond to the service query requirements raised by the user 102.

[0082] For example, the retrieval enhancement method based on the Knowledge Graph builds a knowledge graph index structure by performing entity recognition, relationship extraction and other steps on pre-stored knowledge base files; determines the associated entity nodes based on the similarity between the user query and the entity node description vector, and then mines related information to provide reference information for the large language model.

[0083] For example, the retrieval enhancement method based on hybrid retrieval and rerank combines two or three algorithms from vector retrieval, keyword retrieval and knowledge graph retrieval to implement the hybrid retrieval and rerank strategies.

[0084] For example, the retrieval enhancement method based on intent recognition builds a scene knowledge base index based on the high-frequency vocabulary of the document. During retrieval, the intent of the question is first identified, the corresponding scene knowledge base is obtained, and then text retrieval is performed.

[0085] In combination with the above scenario, it can be seen that in the prior art, by using a large language model, query texts such as business query requirements input by users are processed to generate response information. However, there is a technical problem that the knowledge of the large language model is not updated in a timely manner, which in turn affects the accuracy of the text processing results based on the large language model.

[0086] The text processing method provided by the present application performs multi-level retrieval processing on a cluster summary tree according to a query text input by a user to obtain nodes in the cluster summary tree that are associated with the query text, and generates reference information for the user according to the query text and the business knowledge text in the nodes associated with the query text to perform natural language processing to obtain the business knowledge required by the user, thereby solving the technical problem of low accuracy of text processing results based on a large language model.

[0087] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0088] Figure 2 A flowchart of a text processing method provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the method includes:

[0089] 201. Obtain a query text input by a user, and obtain a cluster summary tree; wherein the query text represents the user's business knowledge query demand; wherein the cluster summary tree includes at least one node; and the node stores the business knowledge text.

[0090] Exemplarily, the execution subject of this embodiment may be an electronic device. For ease of description, the electronic device is hereinafter referred to as the device. The device may be a physical device or a virtualized device that executes the text processing method. When a user performs a business knowledge query based on the device, a query text representing the user's business knowledge query needs may be input into the device. After the device obtains the user's query text, it obtains a pre-stored cluster summary tree, which includes one or more nodes, each of which stores a business knowledge text, so that the cluster summary tree completes the storage of business-related business knowledge texts.

[0091] 202. Perform multi-level search processing on the cluster summary tree according to the query text to obtain at least one target node corresponding to the query text; wherein the target node is a node in the cluster summary tree associated with the query text.

[0092] Exemplarily, the device performs multi-level retrieval processing on the tree storage structure of the cluster summary tree based on the user's query text and the Retrieval-Augmented Generation (RAG) scheme proposed in this application, that is, a preliminary search is first performed to obtain a part of the target nodes, and then further retrieval is performed based on the part of the target nodes to obtain the remaining target nodes.

[0093] For example, first retrieve the business knowledge text stored in each leaf node in the cluster summary tree to obtain each node that has a preliminary association with the query text, such as the nodes where the business knowledge text is related to the text topic or the business scenario to which the text belongs, which are the partial target nodes. Then, based on these partial target nodes, search from the bottom up in the cluster summary tree to further obtain nodes related to the partial target nodes, such as all ancestor nodes of the partial target nodes, which are the remaining target nodes.

[0094] 203. Generate user reference information based on the query text and the business knowledge text of each target node; wherein the reference information represents the business query demand of the user.

[0095] Exemplarily, the device obtains the business knowledge texts stored in all target nodes based on the determined target nodes, and integrates the query text and the business knowledge texts stored in all target nodes, such as integrating the query text, prompt words and the business knowledge texts corresponding to all the acquired target nodes to obtain reference information and related background knowledge that represent the user's business query needs as the final reference information to be processed.

[0096] 204. Perform language processing on the reference information to obtain the user's response information; wherein the response information includes the business knowledge required by the user.

[0097] Exemplarily, the device can be based on natural language processing technology, such as calling a preset large language model, inputting the obtained reference information into the large language model, and using the summary and reasoning capabilities of the large language model to perform language processing on the incoming reference information to generate an answer, that is, the user's reply information, which includes the business knowledge required by the user's query.

[0098] In this embodiment, a text processing method is provided, which is a progressive multi-level retrieval method based on a cluster summary tree. During the retrieval process, the associated partial target nodes are first obtained from a part of the nodes, and then the remaining partial target nodes are gradually obtained to capture a wide range of context information in the cluster summary tree. This method not only improves the accuracy of the retrieval stage, but also supplements rich background information, showing advantages in dealing with complex, multi-step, long text understanding and other problems, and obtaining more accurate text processing results.

[0099] Figure 3 A flowchart of another text processing method provided in an embodiment of the present application is shown as follows: Figure 3 As shown, the method includes:

[0100] 301. Obtain the query text input by the user.

[0101] Exemplarily, this step may refer to step 201 and will not be described in detail here.

[0102] 302. Acquire a business document, and segment the business document to obtain at least one slice corresponding to the business document; wherein the slice represents a business knowledge text.

[0103] For example, Figure 4 A data processing and storage flow chart provided in an embodiment of the present application, such as Figure 4 As shown, the device first obtains the uploaded business document, which may include business knowledge within the entire business scope; the uploaded document is segmented and pre-processed to obtain multiple segmented slices (Chunks), that is, text fragments representing business knowledge texts, to be processed.

[0104] 303. Generate a cluster summary tree based on each slice.

[0105] Exemplarily, the device can store each slice in each node for all the slices, and associate the corresponding nodes according to the relationship between the slices, so as to obtain a cluster summary tree storing all the slices, thereby completing text parsing and storage.

[0106] In one example, step 303 includes the following steps:

[0107] The first step of step 303 is to generate a leaf node corresponding to the slice; wherein the leaf node stores the business knowledge text represented by the slice.

[0108] The second step of step 303 is to perform clustering processing on each slice to obtain at least one first cluster; wherein each slice in the first cluster has similar semantics.

[0109] The third step of step 303 is to process each first cluster to obtain a parent node set; wherein the parent node set includes at least one parent node; the parent node stores the business knowledge text represented by the first cluster.

[0110] Among them, each leaf node and each parent node constitute a cluster summary tree.

[0111] Exemplarily, the device generates a corresponding leaf node for each of all the slices, and stores the business knowledge text represented by each slice in each corresponding leaf node. Based on the text clustering method, the device clusters the slices with similar semantics or matching themes of the business knowledge texts, wherein the number of clustering is a preset number, and multiple cluster cluster texts are obtained, namely the first cluster clusters; wherein the slices in each first cluster cluster are similar in semantics. The number of clustering is determined by the content of the document itself. When vector clustering is performed on each layer of nodes, the optimal number of clusters is first determined based on the Bayesian Information Criterion, and clustering is terminated when the optimal number of clusters at the current level is 0. According to each first cluster, based on the tree storage structure generation method, one or more parent nodes are generated, that is, a parent node set is obtained, wherein each parent node stores the business knowledge text represented by each first cluster; wherein all parent nodes and all leaf nodes constitute a cluster summary tree of the tree storage structure, and these parent nodes are the parent nodes of all leaf nodes, that is, the leaf nodes are low-level nodes, and the parent nodes are high-level nodes, so as to ensure the association relationship and hierarchical relationship between each node, which is convenient for subsequent multi-level retrieval processing.

[0112] In one example, the second step of step 303 includes:

[0113] Step 1: Determine the vector distance between every two slice vectors in the preset vector space among the slice vectors corresponding to each slice.

[0114] Step 2: Perform vector clustering processing on each slice vector according to each vector distance to obtain each first cluster.

[0115] Specifically, combined Figure 4 In the text clustering process, the device performs vectorization based on the word embedding model, completes the embedding of the text fragment corresponding to the slice into the vector, forms its semantic representation in the preset vector space, obtains the slice vector corresponding to each slice, clusters the slice vectors according to the distance in the vector space, and obtains multiple semantically similar clusters, namely the first cluster cluster.

[0116] For example, soft clustering of embedding vectors via a Gaussian mixture model allows nodes to belong to multiple clusters, providing more flexibility to accommodate situations where a single text snippet often contains information related to a variety of topics.

[0117] In one example, the third step of step 303 includes:

[0118] Step 1: Perform language processing on each first cluster to obtain a summary set corresponding to each first cluster; wherein the summary set includes a summary corresponding to each first cluster; and the summary is a text summary of each business knowledge text.

[0119] Step 2: Generate a parent node set based on each abstract in the abstract set.

[0120] Specifically, combined Figure 4 In the process of generating a cluster summary tree, after generating multiple leaf nodes, the text content of each first cluster can be processed based on the summarization ability of the preset large language model to generate a summary of each first cluster, that is, a text summary of each business knowledge text, and a summary set is obtained. According to each summary, one or more parent nodes are generated based on the tree storage structure generation method, that is, a parent node set is obtained, wherein each parent node stores the business knowledge text represented by each summary. Based on the initial slicing and recursive clustering, a tree structure from low level to high level is constructed. The low-level nodes have more accurate detail information, and the high-level nodes summarize a wider range of semantic information. Executing queries and retrievals at different levels can achieve a balance between retrieval accuracy and recall rate.

[0121] In one example, step 2 of the third step of step 303 includes:

[0122] According to the i-th summary set, an i-th layer node set is generated; wherein the node set includes the parent node corresponding to each summary in the summary set.

[0123] Vector clustering is performed on the summary vectors corresponding to each summary in the i-th summary set to obtain the i-th cluster set; wherein the i-th cluster set includes at least one second cluster; and each summary in the second cluster has similar semantics.

[0124] Language processing is performed on each second cluster in the i-th cluster set to obtain an i+1-th summary set; wherein the i+1-th summary set includes the summary corresponding to each second cluster.

[0125] According to the i+1th summary set, the i+1th layer node set is generated; wherein each node set constitutes a parent node set.

[0126] Specifically, combined Figure 4, based on the clustering algorithm, multiple clustering is required. The more clustering times, the richer the semantic information covered by the summary, and the more it can reflect the global semantics. For the i-th summary set obtained after the i-th clustering, the device can generate the parent node corresponding to each summary in the i-th summary set, that is, the i-th layer node set; vectorize each summary in the i-th summary set to obtain the summary vector corresponding to each summary in the i-th summary set, based on the vector clustering algorithm, such as using the distance between the two vectors in the preset vector space in the summary vector corresponding to each summary, perform vector clustering processing on the summary vector corresponding to each summary in the i-th summary set to obtain the i-th cluster cluster set; wherein the i-th cluster cluster set includes one or more second cluster clusters; each summary in each second cluster cluster has similar semantics; based on the summarization ability of the preset large language model, the text content of each second cluster cluster in the i-th cluster cluster set can be processed to generate the i+1-th summary set, and the i+1-th summary set includes the summary of each second cluster cluster in the i-th cluster cluster set. According to each summary in the i+1th summary set, based on the tree storage structure generation method, the i+1th layer node set is generated, that is, a node set of a higher level than the nodes in the i-th layer node set; the node set includes one or more parent nodes, and each summary is stored in the corresponding parent node; each node set constitutes a parent node set. Then, by first splitting, then continuously performing vectorization, clustering, and text summarization, a cluster summary tree is generated to complete text parsing and storage. It not only retains the original text information, but also generates background knowledge based on semantic similarity by integrating related text content, reflecting global semantics layer by layer.

[0127] For example, Figure 5 A schematic diagram of generating a cluster summary tree is provided in an embodiment of the present application, such as Figure 5 As shown in the figure, (1) leaf nodes are generated based on the slices obtained by text segmentation, including 1, 2, 3, 4, 5, and 6; (2) the node content is vectorized, and node clusters are obtained based on the clustering algorithm, such as the node cluster containing 1 and 3, the node cluster containing 2, 4, and 5, and the node cluster containing 2 and 6; (3) based on the large language model, the text content of the node cluster is summarized, and the generated summary is used as the parent node, which is also a layer of nodes, including 7, 8, and 9; 7 and 8 are clustered to obtain node clusters, and 8 and 9 are clustered. Clustering is performed to obtain node clusters. Based on the large language model, the text content of the node clusters is summarized and the generated summary is used as the parent node, which is also the second-level node, including 10 and 11. Steps (2) and (3) are repeated to construct the nodes of each level of the tree layer by layer until vector clustering cannot be performed and the root node root is generated. The final cluster summary tree is generated based on the nodes of each level. Each node stores the current node number, current node level, parent node number, text content, and the vector corresponding to the text content.

[0128] 304. Determine at least one starting node corresponding to the query text from each leaf node in the cluster summary tree.

[0129] Each node in the cluster summary tree includes at least one leaf node and at least one parent node.

[0130] Exemplarily, each node in the cluster summary tree acquired by the device includes at least one leaf node and at least one parent node. Figure 6 A question-answering flow chart based on a cluster summary tree is provided in an embodiment of the present application, such as Figure 6 As shown, the device obtains the user's query text, searches each leaf node in the cluster summary tree, determines the business knowledge text with the highest correlation with the query text, and determines the leaf node where the business knowledge text is located as the starting node.

[0131] For example, embedding is performed based on the same encoding embedding model used to construct the cluster summary tree to obtain a query vector. According to the query vector, a hybrid retriever is established according to multiple preset retrieval methods. According to the query vector, a hybrid search is performed on each leaf node in the cluster summary tree to determine the starting node.

[0132] In one example, step 304 includes the following steps:

[0133] The first step of step 304 is to perform keyword search processing on each leaf node according to the query text to obtain the first candidate document corresponding to the query text; wherein the first candidate document includes at least one business knowledge text in the leaf node that matches the query text.

[0134] The second step of step 304 is to perform semantic retrieval processing on each leaf node according to the query text to obtain a second candidate document corresponding to the query text; wherein the second candidate document includes at least one business knowledge text in a leaf node that is semantically similar to the query text.

[0135] The third step of step 304 is to determine each starting node from each leaf node according to the first candidate document and the second candidate document.

[0136] For example, in combination Figure 6When performing leaf node layer retrieval, the device establishes a hybrid retriever based on two modes: full-text retrieval based on keywords and vector retrieval based on semantic similarity. According to the query text, based on the full-text retrieval method of keywords, keyword retrieval processing is performed on each leaf node to obtain the keyword matching degree between each leaf node and the query text. According to each keyword matching degree, business knowledge texts in multiple leaf nodes with high keyword matching degrees are screened out to form the first candidate document corresponding to the query text; based on the vector retrieval method of semantic similarity, semantic retrieval processing is performed on each leaf node to obtain the semantic similarity between each leaf node and the query text. According to each semantic similarity, business knowledge texts in multiple leaf nodes with high semantic similarity are screened out to obtain the second candidate document corresponding to the query text; based on the two second candidate documents, selection and sorting processing is performed to obtain the best candidate document, which includes one or more business knowledge texts associated with the query text. Each leaf node corresponding to the best candidate document is the starting node. Furthermore, by performing vector retrieval based on low-level text units and locking the location of relevant information, low-level node retrieval can match text with more precise semantic information and avoid inefficient retrieval in a large amount of irrelevant text.

[0137] In one example, the second step of step 304 includes:

[0138] Step 1: Determine the cosine similarity between the query vector corresponding to the query text and the node vector in each leaf node; wherein the node vector is the vector corresponding to the business knowledge text in the leaf node.

[0139] Step 2: Determine the second candidate document based on each cosine similarity.

[0140] Specifically, based on the vector retrieval method of semantic similarity, the device can first determine the vector corresponding to the business knowledge text in each leaf node, that is, the node vector. And calculate the cosine similarity between the query vector corresponding to the query text and each node vector, and select multiple corresponding business knowledge texts with high cosine similarity as the second candidate documents according to the high and low ranking results of each cosine similarity.

[0141] In one example, the third step of step 304 includes:

[0142] Step 1, determine the first relevant information between the first candidate document and the query text; and determine the second relevant information between the second candidate document and the query text; wherein the first relevant information represents the correlation between each business knowledge text in the first candidate document and the query text; and the second relevant information represents the correlation between each business knowledge text in the second candidate document and the query text.

[0143] Step 2: Determine the best candidate document based on the first relevant information and the second relevant information; wherein the best candidate document includes at least one business knowledge text associated with the query text.

[0144] Step 3: Determine the leaf node corresponding to each business knowledge text in the best candidate document from among the leaf nodes as the starting node.

[0145] Specifically, based on the rerank model, the correlation between each business knowledge text and the query text in the first candidate document is determined, that is, the first related information between the first candidate document and the query text; and the correlation between each business knowledge text and the query text in the second candidate document is determined, that is, the second related information between the second candidate document and the query text; the rerank model reorders the two groups of candidate documents according to the correlation between each business knowledge text and the query text to obtain the best candidate document, including one or more business knowledge texts with high correlation, and the leaf node corresponding to each business knowledge text in the best candidate document is used as the starting node.

[0146] 305. According to each starting node, recursively search each parent node in the cluster summary tree to obtain each target node.

[0147] For example, in combination Figure 6 , taking the leaf node corresponding to the best candidate document as the starting node, recursively search each parent node in the cluster summary tree, that is, continue to obtain parent nodes from higher levels until the root node root. For example, for each starting node, obtain the parent node of the starting node, and determine whether the parent node is the root node. If not, continue to obtain the parent node of the parent node until the current node is obtained as the root node. If the current parent node is the root node, all the obtained parent nodes, root nodes and starting nodes are determined as target nodes.

[0148] For example, Figure 7 A schematic diagram of a bottom-up progressive multi-level search method provided in an embodiment of the present application is shown in FIG. Figure 7 As shown in the figure, after query embedding is completed and query vector is obtained, hybrid retrieval and refined ranking are performed at the leaf node layer to obtain the best candidate document and its corresponding leaf node numbered 2, and the parent node of leaf node 2 is obtained to obtain the first-layer nodes numbered 8 and 9; the parent nodes of the first-layer nodes 8 and 9 are obtained respectively to obtain the same second-layer node numbered 11; the parent node of the second-layer node 11 is the root node, and the acquisition layer by layer is stopped; nodes 2, 8, 9, and 11 are nodes related to the query, that is, the final retrieval results.

[0149] In one example, the method further includes: if it is determined that there is no target node corresponding to the query text in the cluster summary tree, generating preset default response information.

[0150] For example, in combination Figure 6 If the device finds that there is no target node corresponding to the query text in the cluster summary tree after searching the cluster summary tree, that is, no relevant document is obtained, a preset default response information is generated and the preset default response information is directly returned to the user.

[0151] 306. Generate user reference information based on the query text and the business knowledge text of each target node; wherein the reference information represents the business query demand of the user.

[0152] Exemplarily, this step may refer to step 203 and will not be described in detail here.

[0153] 307. Perform language processing on the reference information to obtain the user's response information; wherein the response information includes the business knowledge required by the user.

[0154] Exemplarily, this step may refer to step 204 and will not be described in detail here.

[0155] In this embodiment, based on the above embodiments, on the one hand, clustering, summarization and knowledge base establishment are performed based on text semantics, which solves the problem that traditional RAG retrieval methods are difficult to capture global semantics in complex queries, thereby performing well in processing complex and summary reasoning tasks and improving the accuracy of text processing results; on the other hand, bottom-up progressive retrieval ensures an accurate retrieval scope, avoids inefficient retrieval in a large amount of irrelevant text, and reduces time overhead.

[0156] Figure 8 A structural diagram of a text processing device provided in an embodiment of the present application is shown in FIG. Figure 8 As shown, the device comprises:

[0157] The acquisition module 401 is used to acquire the query text input by the user and acquire the cluster summary tree; wherein the query text represents the business knowledge query demand of the user; wherein the cluster summary tree includes at least one node; and the business knowledge text is stored in the node;

[0158] The search module 402 is used to perform multi-level search processing on the cluster summary tree according to the query text to obtain at least one target node corresponding to the query text; wherein the target node is a node in the cluster summary tree associated with the query text;

[0159] The generating module 403 is used to generate user reference information according to the query text and the business knowledge text of each target node; wherein the reference information represents the business query demand of the user;

[0160] The reply module 404 is used to perform language processing on the reference information to obtain the user's reply information; wherein the reply information includes the business knowledge required by the user.

[0161] In a possible implementation, the acquisition module 401 is specifically used to: acquire a business document, and segment the business document to obtain at least one slice corresponding to the business document; wherein the slice represents the business knowledge text; and generate a cluster summary tree based on each slice.

[0162] In one possible implementation, the acquisition module 401 is specifically used to: generate leaf nodes corresponding to slices; wherein the leaf nodes store business knowledge text represented by the slices; cluster each slice to obtain at least one first cluster cluster; wherein the slices in the first cluster cluster have similar semantics; process each first cluster cluster to obtain a parent node set; wherein the parent node set includes at least one parent node; the parent node stores the business knowledge text represented by the first cluster cluster; wherein each leaf node and each parent node constitute a cluster summary tree.

[0163] In a possible implementation, the acquisition module 401 is specifically used to: determine the vector distance between every two slice vectors in the preset vector space among the slice vectors corresponding to each slice; and perform vector clustering processing on each slice vector according to each vector distance to obtain each first cluster cluster.

[0164] In a possible implementation, the acquisition module 401 is further specifically used to: perform language processing on each first cluster to obtain a summary set corresponding to each first cluster; wherein the summary set includes a summary corresponding to each first cluster; the summary is a text summary of each business knowledge text; and generate a parent node set based on each summary in the summary set.

[0165] In a possible implementation, the acquisition module 401 is further specifically used to: generate an i-th layer node set according to the i-th summary set; wherein the node set includes the parent node corresponding to each summary in the summary set; perform vector clustering processing on the summary vectors corresponding to each summary in the i-th summary set to obtain the i-th cluster cluster set; wherein the i-th cluster cluster set includes at least one second cluster cluster; the summaries in the second cluster clusters are semantically similar; perform language processing on each second cluster cluster in the i-th cluster cluster set to obtain the i+1-th summary set; wherein the i+1-th summary set includes the summaries corresponding to each second cluster cluster; generate an i+1-th layer node set according to the i+1-th summary set; wherein each node set constitutes a parent node set.

[0166] In a possible implementation, the cluster summary tree includes at least one leaf node and at least one parent node; the retrieval module 402 is specifically used to: determine at least one starting node corresponding to the query text from each leaf node in the cluster summary tree; and perform recursive retrieval processing on each parent node in the cluster summary tree according to each starting node to obtain each target node.

[0167] In one possible implementation, the retrieval module 402 is specifically used to: perform keyword retrieval processing on each leaf node according to the query text to obtain a first candidate document corresponding to the query text; wherein the first candidate document includes at least one business knowledge text in a leaf node that matches the query text; perform semantic retrieval processing on each leaf node according to the query text to obtain a second candidate document corresponding to the query text; wherein the second candidate document includes at least one business knowledge text in a leaf node that is semantically similar to the query text; and determine each starting node from each leaf node according to the first candidate document and the second candidate document.

[0168] In one possible implementation, the retrieval module 402 is specifically used to: determine the cosine similarity between the query vector corresponding to the query text and the node vector in each leaf node; wherein the node vector is the vector corresponding to the business knowledge text in the leaf node; and determine the second candidate document based on each cosine similarity.

[0169] In a possible implementation, the retrieval module 402 is also specifically used to: determine the first relevant information between the first candidate document and the query text; and determine the second relevant information between the second candidate document and the query text; wherein the first relevant information represents the correlation between each business knowledge text in the first candidate document and the query text; the second relevant information represents the correlation between each business knowledge text in the second candidate document and the query text; determine the best candidate document based on the first relevant information and the second relevant information; wherein the best candidate document includes at least one business knowledge text associated with the query text; and determine from each leaf node the leaf node corresponding to each business knowledge text in the best candidate document as the starting node.

[0170] In a possible implementation manner, the device is further configured to: if it is determined that there is no target node corresponding to the query text in the cluster summary tree, generate preset default response information.

[0171] The device of this embodiment can execute the technical solution in the above method. Its specific implementation process and technical principles are the same and will not be repeated here.

[0172] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application is shown in FIG. Fig. 9As shown, the electronic device includes: a memory 501 and a processor 502; the memory 501 is a memory for storing instructions executable by the processor 502.

[0173] The processor 502 is configured to execute the method provided in the above embodiment.

[0174] The electronic device 500 further includes a receiver 503 and a transmitter 504. The receiver 503 is used to receive instructions and data sent by other devices, and the transmitter 504 is used to send instructions and data to external devices.

[0175] The specific implementation process of the processor can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0176] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0177] An embodiment of the present application also provides a chip for executing instructions, which is used to execute the technical solution of the processing method in the above embodiment.

[0178] The embodiment of the present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed on a computer, the computer executes the technical solution of the processing method of the above embodiment.

[0179] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory, electrically erasable programmable read-only memory, erasable programmable read-only memory, programmable read-only memory, read-only memory, magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0180] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in a dedicated integrated circuit. Of course, the processor and the readable storage medium can also exist in a device as discrete components.

[0181] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, the technical solution of the processing method in the above embodiment can be implemented.

[0182] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0183] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0184] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0185] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0186] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0187] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0188] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A text processing method, characterized in that: include: Obtaining a query text input by a user and obtaining a cluster summary tree; wherein the query text represents the user's business knowledge query demand; wherein the cluster summary tree includes at least one node; and the node stores the business knowledge text; According to the query text, the cluster summary tree is subjected to multi-level retrieval processing to obtain at least one target node corresponding to the query text; wherein the target node is a node in the cluster summary tree associated with the query text; Reference information generated according to the query text and the business knowledge text of each target node; wherein the reference information represents the business query demand of the user; The reference information is language processed to obtain the user's reply information; wherein the reply information includes the business knowledge required by the user.

2. The method according to claim 1, characterized in that Get a cluster summary tree, including: Acquire a business document, and segment the business document to obtain at least one slice corresponding to the business document; wherein the slice represents a business knowledge text; The cluster summary tree is generated according to each of the slices.

3. The method according to claim 2, characterized in that Generating the cluster summary tree according to each of the slices includes: Generate a leaf node corresponding to the slice; wherein the leaf node stores the business knowledge text represented by the slice; Performing clustering processing on each of the slices to obtain at least one first cluster; wherein the slices in the first cluster have similar semantics; Processing each of the first clusters to obtain a parent node set; wherein the parent node set includes at least one parent node; the parent node stores the business knowledge text represented by the first cluster; Wherein, each leaf node and each parent node constitute the cluster summary tree.

4. The method according to claim 3, characterized in that Clustering is performed on each of the slices to obtain at least one first cluster, including: Determine the vector distance between every two slice vectors in the preset vector space among the slice vectors corresponding to each of the slices; According to each of the vector distances, vector clustering processing is performed on each of the slice vectors to obtain each of the first clusters.

5. The method according to claim 3, characterized in that: Each of the first clusters is processed to obtain a parent node set, including: Performing language processing on each of the first clusters to obtain a summary set corresponding to each of the first clusters; wherein the summary set includes a summary corresponding to each of the first clusters; and the summary is a text summary of each of the business knowledge texts; The parent node set is generated according to each summary in the summary set.

6. The method according to claim 5, characterized in that Generating the parent node set according to each abstract in the abstract set includes: Generate an i-th layer node set according to the i-th summary set; wherein the node set includes the parent node corresponding to each summary in the summary set; Performing vector clustering processing on the abstract vectors corresponding to the abstracts in the i-th abstract set to obtain the i-th cluster set; wherein the i-th cluster set includes at least one second cluster; and the abstracts in the second cluster have similar semantics; Performing language processing on each second cluster in the i-th cluster set to obtain an i+1-th summary set; wherein the i+1-th summary set includes a summary corresponding to each second cluster; According to the i+1th summary set, an i+1th layer node set is generated; wherein each of the node sets constitutes the parent node set.

7. The method according to claim 1, characterized in that The cluster summary tree includes at least one leaf node and at least one parent node; according to the query text, the cluster summary tree is subjected to multi-level retrieval processing to obtain at least one target node corresponding to the query text, including: Determine at least one starting node corresponding to the query text from each leaf node in the cluster summary tree; According to each of the starting nodes, a recursive search process is performed on each of the parent nodes in the cluster summary tree to obtain each of the target nodes.

8. The method according to claim 7, characterized in that Determining at least one starting node corresponding to the query text from each leaf node in the cluster summary tree includes: According to the query text, keyword retrieval processing is performed on each leaf node to obtain a first candidate document corresponding to the query text; wherein the first candidate document includes at least one business knowledge text in a leaf node that matches the query text; According to the query text, semantic retrieval processing is performed on each leaf node to obtain a second candidate document corresponding to the query text; wherein the second candidate document includes at least one business knowledge text in a leaf node that is semantically similar to the query text; According to the first candidate document and the second candidate document, each of the starting nodes is determined from each of the leaf nodes.

9. The method according to claim 8, characterized in that According to the query text, semantic retrieval processing is performed on each leaf node to obtain a second candidate document corresponding to the query text, including: Determine the cosine similarity between the query vector corresponding to the query text and the node vector in each leaf node; wherein the node vector is the vector corresponding to the business knowledge text in the leaf node; The second candidate document is determined according to each of the cosine similarities.

10. The method according to claim 8, characterized in that Determining each of the starting nodes from each of the leaf nodes according to the first candidate document and the second candidate document includes: Determine first relevant information between the first candidate document and the query text; and determine second relevant information between the second candidate document and the query text; wherein the first relevant information represents the correlation between each business knowledge text in the first candidate document and the query text; and the second relevant information represents the correlation between each business knowledge text in the second candidate document and the query text; Determine the best candidate document according to the first relevant information and the second relevant information; wherein the best candidate document includes at least one business knowledge text associated with the query text; The leaf node corresponding to each business knowledge text in the best candidate document is determined from each leaf node as the starting node.

11. The method according to any one of claims 1 to 10, characterized in that The method further comprises: If it is determined that there is no target node corresponding to the query text in the cluster summary tree, preset default response information is generated.

12. A text processing device, characterized in that: include: An acquisition module is used to acquire a query text input by a user and acquire a cluster summary tree; wherein the query text represents the user's business knowledge query demand; wherein the cluster summary tree includes at least one node; and the node stores the business knowledge text; A retrieval module, configured to perform a multi-level retrieval process on the cluster summary tree according to the query text to obtain at least one target node corresponding to the query text; wherein the target node is a node in the cluster summary tree associated with the query text; A generating module, used to generate reference information of the user according to the query text and the business knowledge text of each target node; wherein the reference information represents the business query demand of the user; The reply module is used to perform language processing on the reference information to obtain the user's reply information; wherein the reply information includes the business knowledge required by the user.

13. An electronic device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 11 when executed by a processor.

15. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 11 when being executed by a processor.

Images