Question answering method and device based on knowledge graph and refrigeration equipment system

Abstract

The application discloses a knowledge graph-based question answering method and device and a refrigeration equipment system, and belongs to the technical field of refrigeration equipment. The method comprises the following steps: acquiring a created knowledge graph and a preset dictionary; generating a main dictionary according to the knowledge graph and the preset dictionary, and generating a world knowledge dictionary according to the knowledge graph; determining an initial word segmentation result of a question text input by a user according to the world knowledge dictionary; and generating reply information of the question text according to the main dictionary, the initial word segmentation result and the knowledge graph. Therefore, when new knowledge is added to the knowledge graph, the effective identification of related user questions can be realized through field self-adaption, and the model training and manual template configuration are not required, and the flexibility is high.

Description

Technical Field

[0001] This application belongs to the field of refrigeration equipment technology, and in particular relates to a question-answering method, device and refrigeration equipment system based on knowledge graph. Background Technology

[0002] The continuous development of technologies such as artificial intelligence has increased the speed of product iteration and updates for enterprises, and the artificial intelligence of home appliances has gradually become a mainstream development trend. Human-like smart refrigerators are no longer limited to the single task of storage in daily family life; they are constantly making breakthroughs in vision and voice, integrating question-and-answer systems into the product to provide comprehensive services to users.

[0003] With the development of information extraction, knowledge graphs, and related technologies, the use of knowledge graph-based question answering technology to enable dialogue functions in smart refrigerators is becoming increasingly common. However, in the practical application of knowledge graph-based question answering technology, "adaptability" remains a challenge. Adaptability refers to the ability of Natural Language Understanding (NLU) algorithms or models to automatically and effectively identify newly added entities, attributes, relation names, and their alternative names when new knowledge is added to the knowledge graph, without retraining or excessive reliance on manual configuration. Furthermore, the NLU algorithm or model should still correctly recognize the user's intent when the user asks a question related to the newly added knowledge. However, existing knowledge graph-based question answering technologies struggle to effectively identify user questions related to newly added knowledge without training or manual configuration, thus failing to meet the adaptation requirement. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a question-answering method, apparatus, and cooling equipment system based on a knowledge graph, which can adaptively and effectively identify relevant user questions when new knowledge is added to the knowledge graph.

[0005] Firstly, this application provides a question-answering method based on knowledge graphs, including:

[0006] Retrieve the created knowledge graph and preset dictionary;

[0007] A main dictionary is generated based on the knowledge graph and the preset dictionary, and a world knowledge dictionary is generated based on the knowledge graph.

[0008] The initial word segmentation result of the user-input question text is determined based on the world knowledge dictionary;

[0009] Based on the main dictionary, the initial word segmentation results, and the knowledge graph, the response information for the question text is generated.

[0010] In some embodiments, generating a main dictionary based on the knowledge graph and the preset dictionary includes:

[0011] Based on the domain-adaptive weight matrix and each entry in the knowledge graph, determine multiple keywords corresponding to the knowledge graph and category labels corresponding to each keyword;

[0012] The preset dictionary is supplemented based on the keywords and category tags to obtain the main dictionary.

[0013] In some embodiments, the steps for creating the knowledge graph include:

[0014] Obtain the set of knowledge points to be entered;

[0015] Create multiple entries corresponding to each knowledge point in the knowledge point set, and calculate the prediction probability and relative position information of each entry in the knowledge point set;

[0016] The predicted probability and the normalized sum of the relative position information of each entry are taken as the first instance entry of the corresponding knowledge point, and the first object contained in the first instance entry is determined. The first object includes at least two of the following: entity, attribute and relation.

[0017] Create the entry corresponding to the first object to obtain the first defined entry;

[0018] The first instance entry and the first definition entry are stored as nodes to create the knowledge graph.

[0019] In some embodiments, the knowledge graph-based question-answering method further includes:

[0020] Retrieve new knowledge points to be entered;

[0021] Multiple entries corresponding to the new knowledge point are created, the loss value of each entry in the new knowledge point is calculated, and the entry corresponding to the loss value that is less than the preset loss value and has the smallest value is taken as the second instance entry. The second object contained in the second instance entry is determined, and the second object includes at least two of the following: entity, attribute and relation.

[0022] Determine whether a second object exists that is different from the first object;

[0023] If it exists, create the corresponding entry for the second object to obtain the second definition entry, and store the second instance entry and the second definition entry as nodes;

[0024] If it does not exist, the second instance entry will be stored as a node to record the new knowledge point in the knowledge graph.

[0025] In some embodiments, the initial word segmentation result includes multiple word segmentation elements, and determining the initial word segmentation result of the user-input question text based on the knowledge dictionary includes:

[0026] Based on the forward maximum full pattern matching algorithm, the question text input by the user is matched with the knowledge dictionary;

[0027] The words and their corresponding parts of speech that successfully match the question text in the knowledge dictionary are combined into a word segmentation element in the form of data pairs. The consecutive characters that do not match in the question text are merged into a string, and the string is used as a word segmentation element.

[0028] In some embodiments, generating at least one response to the question text based on the main dictionary, the initial word segmentation results, and the knowledge graph includes:

[0029] Iterate through the data type of each segmented element in the initial segmentation result;

[0030] Based on the data type, the initial word segmentation result is processed using a target word segmentation tool to obtain candidate word segmentation results;

[0031] The segmentation elements in the candidate segmentation results are merged using the main dictionary to obtain the target segmentation result;

[0032] Based on the target word segmentation results and the knowledge graph, at least one response message is generated for the question text.

[0033] In some embodiments, generating response information for the question text based on the target word segmentation result and the knowledge graph includes:

[0034] Generate a first feature matrix corresponding to the question text based on the target word segmentation results;

[0035] Determine the second feature matrix corresponding to each instance entry in the knowledge graph;

[0036] Based on the first feature matrix and the second feature matrix, the target instance entry is determined from all the instance entries;

[0037] At least one response message is generated based on the target instance entry to generate the question text.

[0038] Secondly, this application provides a question-answering device based on a knowledge graph, comprising:

[0039] The acquisition unit is used to acquire the created knowledge graph and preset dictionary;

[0040] The generation unit is used to generate a main dictionary based on the knowledge graph and the preset dictionary, and to generate a world knowledge dictionary based on the knowledge graph.

[0041] The determining unit is used to determine the initial word segmentation result of the user-input question text based on the world knowledge dictionary;

[0042] The response unit is used to generate at least one response message for the question text based on the main dictionary, the initial word segmentation results, and the knowledge graph.

[0043] In some embodiments, the generation unit is specifically used for:

[0044] Based on the domain-adaptive weight matrix and each entry in the knowledge graph, determine multiple keywords corresponding to the knowledge graph and category labels corresponding to each keyword;

[0045] The preset dictionary is supplemented based on the keywords and category tags to obtain the main dictionary.

[0046] In some embodiments, the knowledge graph-based question-answering device further includes a creation unit for:

[0047] Obtain the set of knowledge points to be entered;

[0048] Create multiple entries corresponding to each knowledge point in the knowledge point set, and calculate the prediction probability and relative position information of each entry in the knowledge point set;

[0049] The predicted probability and the normalized value corresponding to the relative position information of each entry are used, and the entry with the largest normalized value is taken as the first instance entry of the corresponding knowledge point. The first object contained in the first instance entry is determined, and the first object includes at least two of the following: entity, attribute and relation.

[0050] Create the entry corresponding to the first object to obtain the first defined entry;

[0051] The first instance entry and the first definition entry are stored as nodes to create the knowledge graph.

[0052] In some embodiments, the creation unit is further configured to:

[0053] Retrieve new knowledge points to be entered;

[0054] Create multiple entries corresponding to the new knowledge point;

[0055] Calculate the loss value of each entry in the new knowledge point, and take the entry corresponding to the loss value that is less than the preset loss value and has the smallest value as the second instance entry, and determine the second object contained in the second instance entry, wherein the second object includes at least two of the following: entity, attribute and relation;

[0056] Determine whether a second object exists that is different from the first object;

[0057] If it exists, create the corresponding entry for the second object to obtain the second definition entry, and store the second instance entry and the second definition entry as nodes;

[0058] If it does not exist, the second instance entry will be stored as a node to record the new knowledge point in the knowledge graph.

[0059] In some embodiments, the initial word segmentation result includes multiple word segmentation elements, and the determining unit is specifically used for:

[0060] Based on the forward maximum full pattern matching algorithm, the question text input by the user is matched with the knowledge dictionary;

[0061] The words and their corresponding parts of speech that successfully match the question text in the knowledge dictionary are combined into a word segmentation element in the form of data pairs. The consecutive characters that do not match in the question text are merged into a string, and the string is used as a word segmentation element.

[0062] In some embodiments, the response unit is specifically used for:

[0063] Iterate through the data type of each segmented element in the initial segmentation result;

[0064] Based on the data type, the initial word segmentation result is processed using a target word segmentation tool to obtain candidate word segmentation results;

[0065] The segmentation elements in the candidate segmentation results are merged using the main dictionary to obtain the target segmentation result;

[0066] Based on the target word segmentation results and the knowledge graph, at least one response message is generated for the question text.

[0067] In some embodiments, the response unit is specifically used for:

[0068] Generate a first feature matrix corresponding to the question text based on the target word segmentation results;

[0069] Determine the second feature matrix corresponding to each instance entry in the knowledge graph;

[0070] Based on the first feature matrix and the second feature matrix, the target instance entry is determined from all the instance entries;

[0071] At least one response message is generated based on the target instance entry to generate the question text.

[0072] Thirdly, this application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the knowledge graph-based question-answering method described in any of the preceding claims.

[0073] Fourthly, this application provides a refrigeration equipment system, including a refrigeration device and a server connected in communication. The server includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the knowledge graph-based question-answering method described above.

[0074] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the knowledge graph-based question-answering method described in any of the preceding claims.

[0075] The knowledge graph-based question-answering method, apparatus, storage medium, cooling equipment system, and computer program product provided in this application acquire an existing knowledge graph and a preset dictionary; generate a main dictionary based on the knowledge graph and the preset dictionary, and generate a world knowledge dictionary based on the knowledge graph; determine the initial word segmentation result of the user-input question text based on the world knowledge dictionary; and generate at least one response message for the question text based on the main dictionary, the initial word segmentation result, and the knowledge graph. This enables effective identification of relevant user questions through domain-adaptive processing when new knowledge is added to the knowledge graph, without requiring retraining of the model or manual template configuration, thus offering high flexibility. Attached Figure Description

[0076] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0077] Figure 1 This is a schematic diagram of an application scenario for the refrigeration equipment system provided in the embodiments of this application;

[0078] Figure 2 This is a flowchart illustrating the knowledge graph-based question-answering method provided in an embodiment of this application;

[0079] Figure 3This is a schematic diagram of the framework flow of the knowledge graph-based question answering method provided in the embodiments of this application;

[0080] Figure 4 This is another flowchart illustrating the knowledge graph-based question-answering method provided in this application embodiment;

[0081] Figure 5 This is a schematic diagram of the structure of the knowledge graph-based question-answering device provided in an embodiment of this application;

[0082] Figure 6 This is a schematic diagram of the structure of the refrigeration equipment system provided in the embodiments of this application;

[0083] Figure 7 This is a schematic diagram of the hardware structure of the refrigeration equipment system provided in the embodiments of this application. Detailed Implementation

[0084] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0085] This application provides a knowledge graph-based question-answering method, apparatus, storage medium, cooling equipment system, and computer program product.

[0086] Please see Figure 1 and Figure 2 , Figure 1 This is a schematic diagram illustrating an application scenario of the refrigeration equipment system provided in the embodiments of this application. Figure 2 This is a flowchart illustrating the knowledge graph-based question-answering method provided in this application embodiment. This knowledge graph-based question-answering method is applied to a refrigeration equipment system, which may include a refrigeration device 100 and a client 200 corresponding to the refrigeration device 100. The refrigeration device 100 may be a refrigerator, and the client 200 may be a mobile phone / tablet, or an app / web client on the mobile phone / tablet. The refrigeration device 100 and the client 200 can communicate wirelessly via signals such as 5G, 6G, Wi-Fi, wired connection, or Bluetooth. The following description uses a refrigerator as an example to illustrate the refrigeration device 100.

[0087] Continue as Figure 1As shown, the refrigeration device 100 can be a refrigerator with audio and video capture capabilities and a user interface. The refrigerator is equipped with a microphone array 30 for audio capture, a camera 20 for video recording, a speaker 40 for voice interaction with the user, and an interactive screen 10 for text or graphical interaction with the user. The interactive screen 10 can be mounted on the refrigerator door. When the user opens the refrigerator door, the camera 20 records the user's actions, generating video data. The speaker 40 and microphone 30 work together to interact with the user via audio in a question-and-answer format. The client 200, using a mobile phone as an example, allows users to communicate with the refrigerator via text or voice, manage food information within the refrigerator, and control the refrigerator's operating status.

[0088] In addition, the cooling equipment system may also include other external devices, such as external temperature sensors, cameras 20 or microphones 30 and speakers 40 built into other devices, smart speakers, VR / AR, etc. These devices can be connected to the cooling equipment 100 or the client 200 via wireless signals.

[0089] Specifically, in Figure 2 In this context, the knowledge graph-based question-answering method may include the following steps 301-304, wherein:

[0090] 301. Obtain the created knowledge graph and preset dictionary.

[0091] The preset dictionary is a dictionary with words as keys and a list of business tags as values. The list of business tags includes at least one business tag. When one key corresponds to multiple values ​​(i.e., business tags), these business tags are divided according to different business standards. The preset dictionary is roughly represented as follows:

[0092] {word1:[label1,label2...],

[0093] word2:[label1,label2...],

[0094] ...}

[0095] For details, please see Figure 3 , Figure 3 This is a schematic diagram of the framework flow of the knowledge graph-based question answering method provided in this application embodiment. The knowledge graph and preset dictionary files can be stored locally. When the question answering system performs a question answering task, the knowledge graph and preset dictionary can be loaded into memory from the local files, thus improving the speed of real-time parsing. Alternatively, the knowledge graph and preset dictionary can be loaded from a remote database instead of being stored locally, thus eliminating the need for importing and exporting between local files and the remote database.

[0096] In some embodiments, the steps for creating a knowledge graph include:

[0097] Obtain the set of knowledge points to be entered;

[0098] Create multiple entries for each knowledge point in the knowledge point set, and calculate the predicted probability and relative position information of each entry in the knowledge point set;

[0099] Calculate the normalized sum of the predicted probability and relative position information for each entry, and take the entry with the largest normalized sum as the first instance entry of the corresponding knowledge point, and determine the first object contained in the first instance entry, which includes at least two of the following: entity, attribute and relation.

[0100] Create the entry corresponding to the first object to obtain the first defined entry;

[0101] The first instance entry and the first definition entry are stored as nodes to create the knowledge graph.

[0102] In other words, in knowledge graph construction, unlike the usual approach of using triples as the basic unit, we use "entries" as the basic unit. There are multiple methods for calculating the predicted probability of an entry, including conditional probability prediction, direct probability prediction, and indirect probability prediction. For example, for direct probability prediction, an entry classification model is built for classification, and the model's fitting ability is adjusted by setting a pre-defined entry threshold. Entries within the preset threshold range are considered correct entries in the knowledge set, while those outside are incorrect. For conditional probability prediction, conditional probabilities are used to calculate the probability score.

[0103] Relative position information: The position information of the obtained items is constructed into the same vector space, and then the sequence information obtained from the position information is calculated, such as [0,1,.....], which is the relative position information of each item in the knowledge point set.

[0104] Then, the predicted probability and relative position information are mapped to the same three-dimensional or geometric space and summed to obtain a normalized sum.

[0105] For example, suppose we have a knowledge point called "the optimal temperature for the freezer," which can be represented as an entry:

[0106] {ins property: optimal freezer temperature:}

[0107] Insof: Optimal Temperature

[0108] Host: Freezer

[0109] Content: For freezing, it is recommended to set the temperature to -18 degrees Celsius for faster freezing and better preservation. If your refrigerator has a smart mode, it is recommended to turn it on. In smart mode, the refrigerator automatically adjusts the temperature of the refrigerator and freezer compartments according to changes in the ambient temperature.

[0110] In the above entries, "optimal temperature" after `insof` is an attribute in this knowledge point; "freezer" after `host` is the entity class in this knowledge point; the content after `content` is the attribute value; `ins property` means that the entire entry is an instance of a certain attribute, called an instance entry of the attribute. In our concept, a class of attributes (such as "optimal temperature") can also be regarded as a class, and the corresponding attributes and attribute values ​​of a specific class or entity can be regarded as an "instance" of that class of attributes.

[0111] Each object in the above example entries, namely "freezer" and "optimal temperature," needs to be defined in the knowledge graph schema. The schema is used to define the entities, relationships, attribute categories, and basic constraint rules of the knowledge graph. For example, the definition entry for "freezer" can be as follows:

[0112] {node freezer compartment:

[0113] Subof: Refrigerator compartment

[0114] Names: freezer, freezer compartment, freezer layer, freezer space, freezer area, freeze, freezer box, frozen layer

[0115] standard_name: Freezer

[0116] label:space};

[0117] For example, the definition of "optimal temperature" could be as follows:

[0118] {property optimal temperature:}

[0119] Subof: Device parameters

[0120] Names: Optimal temperature, best temperature, most suitable temperature, most suitable temperature

[0121] standard_name: Optimal temperature

[0122] label:prop};

[0123] In this context, `node` and `property` describe the category of an entry, followed by the entry's name. `subof` refers to the parent class of the current class. `names`, `standard_name`, and `label` are attributes that facilitate slot extraction by the Natural Language Understanding (NLU) module. `names` represents the possible synonyms for the current class in natural language, `standard_name` is the one considered the standard name among many `names`, and `label` is a business-related categorization label for writing rules.

[0124] Compared to triples, the above format is more user-friendly for manual editing and facilitates scripting for automated batch generation. Furthermore, the reason why "optimal temperature" is also an attribute but not included in the freezer's definition entry is because:

[0125] 1) NLU module slotting only requires information such as names, standard_name, and label from the schema, and does not need information from the ins properties. If these are not separated, the NLU module will load a lot of redundant information, which will slow down preloading and real-time parsing. This is the main reason.

[0126] 2) After `ins` properties became independent, developers could add attributes of attributes to these entries to increase dimensions. For example, in addition to `content`, they could add `content_summer` to represent the optimal temperature in summer, `content_winter` to represent the optimal temperature in winter, and so on. This is a secondary reason. Similarly, besides `ins` properties, there are also `inslink`, which are instances of relationships, with a similar meaning. For example, suppose there is another knowledge point, "Eggs are suitable for freezing," which could be represented as an entry like:

[0127] Eggs are best stored in the refrigerator: (ins link)

[0128] Insof: Suitable for placing

[0129] link_left: egg

[0130] link_right: Refrigerator

[0131] Eggs are best stored in the refrigerator. The optimal storage temperature for eggs is 5℃~7℃, which means they can be stored in the refrigerator. If you buy eggs in a whole carton, it's best to put them in the refrigerator along with the carton to ensure maximum freshness.

[0132] Here, "ins property" means that the entire entry is an instance of a certain relation, called an instance entry of the relation. "Suitable to place" after insof is the relation in this knowledge point, and "egg" after link_left and "refrigerator" after link_right are the entity classes in this knowledge point.

[0133] Furthermore, the created knowledge graph can be updated in real time; that is, this knowledge graph-based question-answering method also includes:

[0134] Retrieve new knowledge points to be entered;

[0135] Create multiple entries corresponding to the new knowledge point, calculate the loss value of each entry in the new knowledge point, and take the entry corresponding to the loss value that is less than the preset loss value and has the smallest value as the second instance entry, and determine the second object contained in the second instance entry, which includes at least two of the following: entity, attribute and relation;

[0136] Determine whether a second object exists that is different from the first object;

[0137] If it exists, create the corresponding entry for the second object to obtain the second definition entry, and store the second instance entry and the second definition entry as nodes;

[0138] If it does not exist, the second instance entry will be stored as a node to record the new knowledge point in the knowledge graph.

[0139] For newly added knowledge points, corresponding instance entries need to be created. If the objects in the instance entries have already been defined in the schema, then only the newly created instance entries need to be added. If they have not been defined, then the entries of each object need to be entered into the knowledge graph to update the knowledge graph.

[0140] Example of calculating the loss value for a single entry: Assuming a binary logistic regression softmax classifier is used to calculate the loss value, the formula is:

[0141] Among them, f j f represents the j-th entry. yi L represents the actual value of the i-th entry. i Let be the loss value for the i-th entry. The minimum loss value can be adjusted using the gradient.

[0142] 302. Generate a main dictionary based on the knowledge graph and the preset dictionary, and generate a world knowledge dictionary based on the knowledge graph.

[0143] The main dictionary is a dictionary aligned with three types of knowledge: preference and interest knowledge, style knowledge, and common knowledge.

[0144] In some embodiments, knowledge alignment is controlled by a domain-adaptive weight matrix that governs the generation of the main dictionary based on preference interest knowledge, style knowledge, and commonality knowledge. See also Figure 4 , Figure 4 This is another flowchart illustrating the knowledge graph-based question-answering method provided in this application embodiment. Step 302 above, "generating a main dictionary based on the knowledge graph and the preset dictionary," specifically includes:

[0145] 3021. Based on the domain-adaptive weight matrix and each entry in the knowledge graph, determine the multiple keywords corresponding to the knowledge graph and the category label corresponding to each keyword;

[0146] 3022. Supplement the preset dictionary with keywords and category tags to obtain the main dictionary.

[0147] The process begins by aligning three types of knowledge—preference and interest knowledge, style knowledge, and commonality knowledge—in the knowledge graph using a domain-adaptive weight matrix. Then, keywords and category tags are extracted from the aligned knowledge, and the pre-defined dictionary is supplemented. There are various methods to control knowledge alignment using a domain-adaptive weight matrix, including multi-factor models, equal-weighted synthesis methods, and linear regression. For example, a linear regression method could be used: Y = W1X1 + W2X2 + W3X3 + W, where W... 1、 W 2、 W3 represents the knowledge (item) weight variable, and X1, X2, and X3 are knowledge (item) variables. Y is the alignment knowledge variable, and W is the network parameter tuning, i.e., the weight difference.

[0148] Please continue reading Figure 3 The keywords are the contents of items like `names` and `standard_name`, and the category labels are the contents of `label`. Specifically, the NLU module extracts slots from the `names`, `standard_name`, and `label` information in the definition entries of each entity class, attribute class, and relationship class to obtain keywords and category labels. Then, each keyword is treated as a key, and the category label is used as a value to supplement the preset dictionary. The supplemented preset dictionary becomes the main dictionary. Supplementing means that if a corresponding key is not found in the preset dictionary, the keyword and category label are used as a key.<key,value> For newly added keywords, if the corresponding key exists in the preset dictionary but the corresponding value does not contain the label, then the label is added to the value; otherwise, it is ignored. Simultaneously, a separate dedicated vocabulary (i.e., a world knowledge dictionary) is compiled for each keyword listed in the knowledge graph, so that words in the knowledge graph can be prioritized for word segmentation during subsequent parsing.

[0149] 303. Determine the initial word segmentation result of the user-input question text based on the world knowledge dictionary.

[0150] The question text can be text data directly entered by the user into the refrigeration equipment system. For example, the user can enter the question to be answered by directly interacting with the refrigerator by voice or by entering the text input box provided by the mobile APP, mini program or web client. For example, the user can enter the text content "What should patients with hypertension pay attention to in their diet?" It can also be collected by other devices and sent to the refrigeration equipment system through the network. Alternatively, the question text data can also be text data converted from other non-text data, such as text data obtained by recognizing voice data through voice recognition technology.

[0151] In some embodiments, please continue to see Figure 4 The initial word segmentation result includes multiple word segmentation elements, and step 303 above specifically includes:

[0152] 3031. Based on the forward maximum full pattern matching algorithm, match the user-input question text with the world knowledge dictionary;

[0153] 3032. Combine the words and their corresponding parts of speech that successfully match the question text in the world knowledge dictionary into a single word segmentation element in the form of data pairs. Merge consecutive characters that do not match in the question text into a single string and use this string as a word segmentation element.

[0154] In other words, assuming the question text is "What is the best temperature in the freezer?", the initial word segmentation result seq0 obtained by parsing the proprietary world knowledge dictionary based on the forward maximum full pattern matching algorithm is: [("freezer", "n"), ("best temperature", "n"), "what is the temperature?"]. In this seq0, words that can be found in the world knowledge dictionary will form a pair structure (i.e., the structure of data pairs), which consists of words and their corresponding parts of speech (the default part of speech of words in the world knowledge dictionary is n, i.e., noun). Other characters that cannot be found will be continuously merged into a string, such as "what is the temperature?".

[0155] 304. Based on the main dictionary, the initial word segmentation results, and the knowledge graph, generate at least one response to the question text.

[0156] This allows for the generation of the top N highest-scoring responses to a question, where N is a positive integer.

[0157] In some embodiments, step 304 specifically includes:

[0158] 3041. Iterate through the data type of each segmented element in the initial segmentation result;

[0159] 3042. Based on this data type, use the target word segmentation tool to process the initial word segmentation result to obtain candidate word segmentation results;

[0160] 3043. Use the main dictionary to merge the segmented elements in the candidate segmentation results to obtain the target segmentation result;

[0161] 3044. Based on the target word segmentation result and the knowledge graph, generate at least one response to the question text.

[0162] In other words, the initial segmentation result seq0 is further segmented using a target word segmentation tool (such as Jieba, NLTK, etc.). Specifically, a new seq1 is created, and then each segmented element in seq0 is traversed to determine its data type. If it is a pair, it is directly added to seq1; if it is a string, the string is segmented using the target word segmentation tool to obtain a pair sequence, which is then extended into seq1. After completing the traversal, the segmented + part-of-speech tagging sequence seq1 (i.e., the candidate segmentation result) is obtained. Then, the main dictionary is used to supplement the segmentation of seq1, checking which consecutive pairs in seq1 can be merged to obtain the final target word segmentation result seq2.

[0163] For example, for the above seq0, after further segmentation using the target segmentation tool, the candidate segmentation result seq1 is: [(“freezer”, “n”), (“best temperature”, “n”), (“is”, “v”), (“how much”, “m”), (“degree”, “q”), (“ne”, “y”). Then, by merging some segmented elements in seq1 using the main dictionary, the target segmentation result seq2 is: [(“freezer”, “n”), (“best temperature”, “n”), (“is”, “v”), (“how much degree”, “mq”), (“ne”, “y”), which is the final segmentation result of the question text “What is the best temperature of the freezer?”.

[0164] In other embodiments, word segmentation of the problem text involves two steps:

[0165] Step 1: First, use word segmentation tools such as Jieba to generate an original word segmentation result. For example, "I want to eat beer and braised pork", the original word segmentation result is "I / n want / vd eat / v beer / n braised pork / n".

[0166] Step 2: Based on the forward maximum matching algorithm, check which words in the original word segmentation result can be combined into a larger word in the dictionary. For example, in the above example, "beer" and "braised pork in brown sauce" are segmented into two words, but it can be found in the dictionary that "beer-braised pork in brown sauce" is a dish name, so it can be known that these two words can be combined into "beer-braised pork in brown sauce".

[0167] There are two problems in this process that prevent the realization of the adaptability of the NLU module. One is that there is only a dictionary and no knowledge graph. If the information in the knowledge graph needs to be introduced, the words in the knowledge graph need to be manually added to the dictionary. The other is that tools like Jieba are prioritized in this mechanism, but once they make word segmentation errors, it may lead to errors even for words that exist in the dictionary, and it is necessary to write word segmentation correction rules to correct them, which will involve a lot of work. For example, in the sentence "Put sea cucumbers in the refrigerator", "participate" may be segmented as a single word, so even if the forward maximum matching algorithm is used to check the dictionary later, "sea cucumbers" cannot be recognized as an ingredient, resulting in a relatively low word segmentation accuracy and affecting the response effect of the subsequent large model.

[0168] Based on the above, the solution of this application improves the word segmentation process from two aspects:

[0169] 1) Directly introduce the words in the knowledge graph as a proprietary dictionary outside the dictionary, so that the terms in the knowledge graph are automatically synchronized to the parsing process of the NLU module, and the corresponding knowledge allows knowledge embedding and query through some specific grammars.

[0170] 2) In the basic word segmentation part, change the strategy of prioritizing the word segmentation tool and adopt the strategy of prioritizing the terms in the proprietary dictionary to avoid the word segmentation errors of the word segmentation tool affecting the terms in the knowledge graph.

[0171] It is easy to know that the word segmentation process designed in the solution of this application distinguishes the main vocabulary and the proprietary vocabulary of the knowledge graph, and designs the corresponding priority order and word segmentation process, realizing the priority word segmentation of the words in the knowledge graph and avoiding the cascading errors caused by the word segmentation errors of the word segmentation tool. The newly added terms in the knowledge graph can adaptively adjust the word segmentation result, thus finally achieving the adaptive effect, that is, when the knowledge graph changes, without retraining and without relying too much on additional manual configuration, it can automatically and effectively identify the newly added entities, attributes, relationship names, etc. and their aliases. When the user asks relevant questions, the NLU algorithm or model can still correctly identify the user's intention and the entities, attributes, and relationships mentioned in the question.

[0172] In some embodiments, the above step 3044 specifically includes:

[0173] Generate a first feature matrix corresponding to the question text according to the target word segmentation result;

[0174] Determine the second feature matrix corresponding to each instance entry in the knowledge graph;

[0175] Based on the first feature matrix and the second feature matrix, the target instance entry is determined from all the instance entries;

[0176] Generate at least one response message for the question text based on the target instance entry.

[0177] For the target word segmentation result seq2 mentioned above, the business tag of each element can be read from the dictionary, and the corresponding first feature matrix can be as follows:

[0178] Word segmentation Part of speech Pinyin Feature label 1 ... Business Tags freezer n leng3, dong4, shi4 space Optimal temperature n zui4,jia1,de5,wen1,du4 prop yes v shi4 How many degrees mq duo1,shao3,du4 Woolen cloth y ne5 END

[0179] Simultaneously, for each instance entry in the knowledge graph, the NLU module can extract slots from the names, standard_name, and label information in its associated definition entry to obtain the keywords and category labels corresponding to each instance entry. A second feature matrix is ​​then generated using the same method, with one second feature matrix for each instance entry. Next, the keywords and business labels in the first and second feature matrices are matched. If they match, the matched entry is designated as the target instance entry, which is also the search result of the knowledge graph. Finally, a large model can be used to generate the top N highest-scoring responses based on the prompt and the target instance entry.

[0180] As described above, the knowledge graph-based question-answering method provided in this application obtains an existing knowledge graph and a preset dictionary; generates a main dictionary based on the knowledge graph and the preset dictionary, and generates a world knowledge dictionary based on the knowledge graph; determines the initial word segmentation result of the user-input question text based on the world knowledge dictionary; and generates response information for the question text based on the main dictionary, the initial word segmentation result, and the knowledge graph. This allows for effective identification of relevant user questions through domain adaptation when new knowledge is added to the knowledge graph, without requiring retraining of the model or manual template configuration, thus offering high flexibility.

[0181] Based on the methods described in the above embodiments, this application also provides a knowledge graph-based question-answering method applied to a refrigeration equipment system. Please refer to... Figure 5 , Figure 5 This is a schematic diagram of the structure of a knowledge graph-based question-answering device provided in an embodiment of this application. The knowledge graph-based question-answering device 400 includes an acquisition unit 401, a generation unit 402, a determination unit 403, and a response unit 404, wherein:

[0182] Acquisition unit 401 is used to acquire the created knowledge graph and preset dictionary;

[0183] The generation unit 402 is used to generate a main dictionary based on the knowledge graph and the preset dictionary, and to generate a world knowledge dictionary based on the knowledge graph.

[0184] The determining unit 403 is used to determine the initial word segmentation result of the user-input question text based on the world knowledge dictionary;

[0185] The response unit 404 is used to generate at least one response message for the question text based on the main dictionary, the initial word segmentation result, and the knowledge graph.

[0186] In some embodiments, the generation unit 402 is specifically used for:

[0187] Based on the domain-adaptive weight matrix and each entry in the knowledge graph, determine multiple keywords corresponding to the knowledge graph, as well as the category label corresponding to each keyword;

[0188] The main dictionary is obtained by supplementing the preset dictionary with keywords and category tags.

[0189] In some embodiments, the knowledge graph-based question-answering device 400 further includes a creation unit for:

[0190] Retrieve the set of knowledge points to be entered;

[0191] Create multiple entries for each knowledge point in the knowledge point set, and calculate the predicted probability and relative position information of each entry in the knowledge point set;

[0192] Calculate the normalized sum of the predicted probability and relative position information for each entry, and take the entry with the largest normalized sum as the first instance entry of the corresponding knowledge point, and determine the first object contained in the first instance entry, which includes at least two of the following: entity, attribute and relation.

[0193] Create the entry corresponding to the first object to obtain the first defined entry;

[0194] The first instance entry and the first definition entry are stored as nodes to create the knowledge graph.

[0195] In some embodiments, the creation unit is further configured to:

[0196] Retrieve new knowledge points to be entered;

[0197] Multiple entries corresponding to the new knowledge point are created, the loss value of each entry in the new knowledge point is calculated, and the entry corresponding to the loss value that is less than the preset loss value and has the smallest value is taken as the second instance entry. The second object contained in the second instance entry is determined, and the second object includes at least two of the following: entity, attribute and relation.

[0198] Determine whether a second object exists that is different from the first object;

[0199] If it exists, create the corresponding entry for the second object to obtain the second definition entry, and store the second instance entry and the second definition entry as nodes;

[0200] If it does not exist, the second instance entry will be stored as a node to record the new knowledge point in the knowledge graph.

[0201] In some embodiments, the initial word segmentation result includes multiple word segmentation elements, and the determining unit 403 is specifically used for:

[0202] Based on the forward maximum full pattern matching algorithm, the user-input question text is matched with the world knowledge dictionary;

[0203] The words and their corresponding parts of speech that successfully match the question text in the world knowledge dictionary are combined into a single word segmentation element as data pairs. The consecutive characters that do not match in the question text are merged into a single string, and this string is used as a word segmentation element.

[0204] In some embodiments, the response unit 404 is specifically used for:

[0205] Iterate through the data type of each segmented element in the initial segmentation result;

[0206] Based on this data type, the initial segmentation result is processed using a target segmentation tool to obtain candidate segmentation results;

[0207] The segmented elements in the candidate segmentation results are merged using the main dictionary to obtain the target segmentation result;

[0208] Based on the target word segmentation results and the knowledge graph, generate at least one response to the question text.

[0209] In some embodiments, the response unit 404 is specifically used for:

[0210] Generate the first feature matrix corresponding to the problem text based on the target word segmentation result;

[0211] Determine the second feature matrix corresponding to each instance entry in the knowledge graph;

[0212] Based on the first feature matrix and the second feature matrix, the target instance entry is determined from all the instance entries;

[0213] Generate at least one response message for the question text based on the target instance entry.

[0214] It should be noted that the specific details of each module unit in the knowledge graph-based question answering device 400 have been described in detail in the embodiments of the knowledge graph-based question answering method, and will not be repeated here.

[0215] In some embodiments, the knowledge graph-based question-answering device in this application can be a refrigeration equipment system, or a component within the refrigeration equipment system, such as an integrated circuit or a chip. The refrigeration equipment system includes, but is not limited to, refrigerators, freezers, display cases, beverage cabinets, wine cabinets, refrigerated display cases, and refrigerated vending machines, etc., and has diverse structural forms and a wide range of applications.

[0216] In some embodiments, such as Figure 6 As shown, this application embodiment also provides a refrigeration equipment system 500, including a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the program is executed by the processor 501, it implements the various processes of the above-described knowledge graph-based question-answering method embodiment and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0217] It should be noted that the refrigeration equipment system in this application embodiment includes the aforementioned mobile refrigeration equipment system and non-mobile refrigeration equipment system.

[0218] Figure 7 A schematic diagram of the hardware structure of a refrigeration equipment system according to an embodiment of this application.

[0219] The refrigeration equipment system 600 includes, but is not limited to, the following components: radio frequency unit 601, network module 602, audio output unit 603, input unit 604, sensor 605, display unit 606, user input unit 607, interface unit 608, memory 609, and processor 610.

[0220] Those skilled in the art will understand that the refrigeration system 700 may also include a power supply (such as a battery) for supplying power to various components. The power supply may be logically connected to the processor 610 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 7 The structure of the refrigeration equipment system shown in the figure does not constitute a limitation on the refrigeration equipment system. The refrigeration equipment system may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0221] In some embodiments, the processor 610 is configured to:

[0222] Retrieve the created knowledge graph and preset dictionary;

[0223] A main dictionary is generated based on the knowledge graph and the preset dictionary, and a world knowledge dictionary is generated based on the knowledge graph.

[0224] The initial word segmentation results of the user's input question text are determined based on this world knowledge dictionary;

[0225] Based on the main dictionary, the initial word segmentation results, and the knowledge graph, generate at least one response to the question text.

[0226] In some embodiments, the processor 610 is configured to:

[0227] Based on the domain-adaptive weight matrix and each entry in the knowledge graph, determine the multiple keywords corresponding to the knowledge graph and the category label corresponding to each keyword;

[0228] The preset dictionary is supplemented based on the keyword and category tag to obtain the main dictionary.

[0229] In some embodiments, the processor 610 is further configured to:

[0230] Retrieve the set of knowledge points to be entered;

[0231] Create multiple entries for each knowledge point in the knowledge point set, and calculate the predicted probability and relative position information of each entry in the knowledge point set;

[0232] The predicted probability and the normalized value corresponding to the relative position information of each entry are used, and the entry with the largest normalized value is taken as the first instance entry of the corresponding knowledge point. The first object contained in the first instance entry is determined, and the first object includes at least two of the following: entity, attribute and relation.

[0233] Create the entry corresponding to the first object to obtain the first defined entry;

[0234] The first instance entry and the first definition entry are stored as nodes to create the knowledge graph.

[0235] In some embodiments, the processor 610 is further configured to:

[0236] Retrieve new knowledge points to be entered;

[0237] Create multiple entries corresponding to the new knowledge point, calculate the loss value of each entry in the new knowledge point, and take the entry corresponding to the loss value that is less than the preset loss value and has the smallest value as the second instance entry, and determine the second object contained in the second instance entry, which includes at least two of the following: entity, attribute and relation;

[0238] Determine if there exists a second object that is different from the first object;

[0239] If it exists, create the corresponding entry for the second object to obtain the second definition entry, and store the second instance entry and the second definition entry as nodes;

[0240] If it does not exist, the second instance entry will be stored as a node to record the new knowledge point in the knowledge graph.

[0241] In some embodiments, the initial word segmentation result includes multiple word segmentation elements, and the processor 610 is used for:

[0242] Based on the forward maximum full pattern matching algorithm, the question text input by the user is matched with the knowledge dictionary;

[0243] The words and their corresponding parts of speech that successfully match the question text in the knowledge dictionary are combined into a single word segmentation element in the form of data pairs. The consecutive characters that do not match in the question text are merged into a single string, and this string is used as a word segmentation element.

[0244] In some embodiments, the processor 610 is configured to:

[0245] Iterate through the data type of each segmented element in the initial segmentation result;

[0246] Based on this data type, the initial segmentation result is processed using a target segmentation tool to obtain candidate segmentation results;

[0247] The segmented elements in the candidate segmentation results are merged using the main dictionary to obtain the target segmentation result;

[0248] Based on the target word segmentation results and the knowledge graph, generate at least one response to the question text.

[0249] In some embodiments, the processor 610 is configured to:

[0250] Generate the first feature matrix corresponding to the problem text based on the target word segmentation result;

[0251] Determine the second feature matrix corresponding to each instance entry in the knowledge graph;

[0252] Based on the first feature matrix and the second feature matrix, the target instance entry is determined from all instance entries;

[0253] Generate at least one response message for the question text based on the target instance entry.

[0254] It should be understood that, in this embodiment, the input unit 604 may include a graphics processing unit (GPU) 6041 and a microphone 6042. The GPU 6041 processes image data of still images or videos obtained by an image capture device (such as a camera) in video capture mode or image capture mode. The display unit 606 may include a display panel 6061, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, or the like. The user input unit 607 includes at least one of a touch panel 6071 and other input devices 6072. The touch panel 6071 is also called a touch screen. The touch panel 6071 may include two parts: a touch detection device and a touch controller. Other input devices 6072 may include, but are not limited to, a physical keyboard, function keys (such as volume control buttons, power buttons, etc.), a trackball, a mouse, and a joystick, which will not be described in detail here.

[0255] The memory 609 can be used to store software programs and various data. The memory 609 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 609 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory 609 in this embodiment includes, but is not limited to, these and any other suitable types of memory.

[0256] Processor 610 may include one or more processing units; processor 610 integrates an application processor and a modem processor, wherein the application processor mainly handles operations involving the operating system, user interface, and applications, and the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into processor 610.

[0257] This application also provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described knowledge graph-based question-answering method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0258] The processor is the processor in the refrigeration device system described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0259] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described knowledge graph-based question-answering method.

[0260] The processor is the processor in the refrigeration device system described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0261] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0262] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0263] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0264] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0265] In the description of this application, "multiple" means two or more.

[0266] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0267] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A question-answering method based on knowledge graphs, characterized in that, include: Retrieve the created knowledge graph and preset dictionary; A main dictionary is generated based on the knowledge graph and the preset dictionary, and a world knowledge dictionary is generated based on the knowledge graph. The initial word segmentation result of the user-input question text is determined based on the world knowledge dictionary; Based on the main dictionary, the initial word segmentation results, and the knowledge graph, at least one response message is generated for the question text.

2. The question-answering method based on knowledge graphs according to claim 1, characterized in that, The step of generating the main dictionary based on the knowledge graph and the preset dictionary includes: Based on the domain-adaptive weight matrix and each entry in the knowledge graph, determine multiple keywords corresponding to the knowledge graph and category labels corresponding to each keyword; The preset dictionary is supplemented based on the keywords and category tags to obtain the main dictionary.

3. The question-answering method based on knowledge graphs according to claim 1, characterized in that, The steps for creating the knowledge graph include: Obtain the set of knowledge points to be entered; Create multiple entries corresponding to each knowledge point in the knowledge point set, and calculate the prediction probability and relative position information of each entry in the knowledge point set; Calculate the predicted probability and the normalized sum of the relative position information for each entry, and take the entry with the largest normalized sum as the first instance entry of the corresponding knowledge point, and determine the first object contained in the first instance entry, wherein the first object includes at least two of entities, attributes and relations. Create the entry corresponding to the first object to obtain the first defined entry; The first instance entry and the first definition entry are stored as nodes to create the knowledge graph.

4. The question-answering method based on knowledge graphs according to claim 3, characterized in that, Also includes: Retrieve new knowledge points to be entered; Create multiple entries corresponding to the new knowledge point; Calculate the loss value of each entry in the new knowledge point, and take the entry corresponding to the loss value that is less than the preset loss value and has the smallest value as the second instance entry, and determine the second object contained in the second instance entry, wherein the second object includes at least two of the following: entity, attribute and relation; Determine whether a second object exists that is different from the first object; If it exists, create the corresponding entry for the second object to obtain the second definition entry, and store the second instance entry and the second definition entry as nodes; If it does not exist, the second instance entry will be stored as a node to record the new knowledge point in the knowledge graph.

5. The question-answering method based on knowledge graphs according to claim 1, characterized in that, The initial word segmentation result includes multiple word segmentation elements. The initial word segmentation result determined based on the world knowledge dictionary for the user-input question text includes: Based on the forward maximum full pattern matching algorithm, the user-input question text is matched with the world knowledge dictionary; The words and their corresponding parts of speech that successfully match the question text in the world knowledge dictionary are combined into a word segmentation element in the form of data pairs. The consecutive characters that do not match in the question text are merged into a string, and the string is used as a word segmentation element.

6. The question-answering method based on knowledge graphs according to claim 1, characterized in that, The step of generating at least one response to the question text based on the main dictionary, the initial word segmentation results, and the knowledge graph includes: Iterate through the data type of each segmented element in the initial segmentation result; Based on the data type, the initial word segmentation result is processed using a target word segmentation tool to obtain candidate word segmentation results; The segmentation elements in the candidate segmentation results are merged using the main dictionary to obtain the target segmentation result; Based on the target word segmentation results and the knowledge graph, at least one response message is generated for the question text.

7. The question-answering method based on knowledge graphs according to claim 6, characterized in that, The step of generating at least one response to the question text based on the target word segmentation result and the knowledge graph includes: Generate a first feature matrix corresponding to the question text based on the target word segmentation results; Determine the second feature matrix corresponding to each instance entry in the knowledge graph; Based on the first feature matrix and the second feature matrix, the target instance entry is determined from all the instance entries; At least one response message is generated based on the target instance entry to generate the question text.

8. A question-answering device based on a knowledge graph, characterized in that, include: The acquisition unit is used to acquire the created knowledge graph and preset dictionary; The generation unit is used to generate a main dictionary based on the knowledge graph and the preset dictionary, and to generate a world knowledge dictionary based on the knowledge graph. The determining unit is used to determine the initial word segmentation result of the user-input question text based on the world knowledge dictionary; The response unit is used to generate at least one response message for the question text based on the main dictionary, the initial word segmentation results, and the knowledge graph.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the knowledge graph-based question-answering method as described in any one of claims 1-7.

10. A refrigeration equipment system, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the knowledge graph-based question-answering method as described in any one of claims 1-7.

Images