An intelligent analysis method and system for standardizing home appliance service order information, an electronic device, and a storage medium

Abstract

The application discloses a kind of home appliance service order information standardization intelligent analysis method, comprising: according to the matching degree and existence confidence of each preset basic entity in order text, determine target basic entity and confidence entity;According to confidence entity, determine corresponding target characteristic entity;According to matching degree, existence confidence and preset reliability threshold, determine reliability entity;According to the cosine similarity between any two effective entities, determine conflict degree;According to the base score of effective entity and conflict degree, determine the modified score of effective entity;Effective entity meeting the condition is used as decision result and outputs conflict early warning and conflict processing information;According to preset label template, generate service prompt label.The application can convert unstructured natural language order into standard service prompt label, and service engineers can intuitively obtain user demands and precautions.The application also discloses a system, electronic equipment and storage medium for implementing the above method.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and in particular to a standardized intelligent parsing method, system, electronic device, and storage medium for home appliance service order information. Background Technology

[0002] In recent years, with the development of artificial intelligence technology, various skills service platforms have widely integrated intelligent customer service systems to facilitate users' access to services. Submitting service requests or placing orders through the intelligent customer service interfaces of these platforms has become a common practice.

[0003] Currently, existing technological solutions in this field typically rely on pre-defined, structured interaction processes. Specifically, intelligent customer service often employs a multiple-choice Q&A guidance strategy. When a user intends to place an order, the system progressively provides a series of options for the user to choose from, such as confirming service type, specifications, and time parameters by clicking a button or replying with a specific number, thereby gradually building a structured, standard order. The advantage of this approach is its clear process and explicit direction, effectively avoiding ambiguous user input and ensuring that the information obtained by the system is standardized and processable.

[0004] However, this existing technological solution has significant limitations. Intelligent customer service systems are essentially just decision trees or state machines operating along fixed paths, unable to understand and process free-form text input by users—that is, natural language orders. For example, when a user directly inputs a natural language description containing multiple elements, such as "I want to book house cleaning for 3 PM tomorrow, two hours, window cleaning," the system can only respond with options like "Please select service type: 1. Appliance repair; 2. House cleaning..." forcing the user to switch from a natural dialogue mode to a mechanical menu selection mode. This approach struggles to fully understand and parse the complex intentions, personalized requirements, and unstructured contextual information contained in the user's free text input. For instance, users may mix multiple elements in their descriptions, such as time, location, specific service requirements, and special preferences, and existing systems typically cannot perform complete semantic extraction, association, and standardization of these elements. Moreover, due to insufficient depth in understanding natural language, the system often fails to accurately capture the user's personalized or implicit needs, resulting in incomplete and inaccurate order information, and the final service plan may deviate from the user's true expectations.

[0005] The shortcomings of this interaction model lead to cumbersome user operations, which actually reduces efficiency, especially when user needs are clear. At the same time, it limits the applicable scenarios for intelligent customer service, making it unable to handle complex, varied, or personalized natural language requests, thus hindering further improvements in the automation level of service platforms. Summary of the Invention

[0006] To address the aforementioned problems in the existing technology, this invention provides a standardized intelligent parsing method, system, electronic device, and storage medium for home appliance service order information. The technical problem to be solved by this invention is achieved through the following technical solution: The first aspect of this invention provides a standardized intelligent parsing method for home appliance service order information, comprising the following steps: The target basic entity is determined based on the matching degree of each preset basic entity in the order text; wherein, the preset basic entities include: brand entity, category entity and basic specification entity; Based on the confidence level of the presence of each preset user demand entity and the target basic entity in the order text, a confident entity is determined; The corresponding target characteristic entity is determined in the preset characteristic entity library based on the confidence entity; Based on the matching degree, the existence confidence degree, and the preset reliability threshold, a reliable entity is determined from the confident entities; The reliability entity and the target characteristic entity are taken as valid entities, and the conflict degree is determined based on the cosine similarity between any two valid entities. The base score for each valid entity is determined based on the reliability score of the valid entity and the preset priority weight; When the conflict degree of the valid entity is greater than the preset conflict threshold, the corrected score of the valid entity is determined based on the base score and conflict degree of the valid entity. The effective entities whose base score and modified score are greater than or equal to a preset threshold are used as the decision results, and conflict warning and conflict handling information are output according to the effective entities whose conflict degree is greater than the preset conflict threshold. Service prompt tags are generated based on the preset tag template, the valid entity, the decision result, the conflict warning, and the conflict handling information.

[0007] In one embodiment of the present invention, determining the target basic entity based on the matching degree of each preset basic entity in the order text includes: The matching degree of each preset basic entity is determined based on the word frequency of each preset basic entity in the order text and the preset basic weight of each preset basic entity; The target basic entity is determined based on the matching degree and the first threshold.

[0008] In one embodiment of the present invention, determining the confidence entity based on the presence confidence of each preset user demand entity and the target basic entity in the order text includes: The presence confidence of each preset user demand entity and the target basic entity in the order text is calculated based on the BERT model. Confidential entities are determined based on the aforementioned confidence level and the second threshold.

[0009] In one embodiment of the present invention, determining a reliable entity from the confident entities based on the matching degree, the existence confidence degree, and a preset reliability threshold includes: Based on the matching degree and the existence confidence degree of the trusted entity, a reliability score is determined for each trusted entity; The reliability entity is determined based on the reliability score and the preset reliability threshold.

[0010] In one embodiment of the present invention, the formula for calculating the existence confidence level is: in, This indicates the confidence level of the presence of each preset user demand entity and the target basic entity in the order text. This represents the order text. This represents the preset user requirement entity and the target basic entity. express and The fused semantic vector express The model's weight matrix, express The bias term of the model, This represents the normalization function.

[0011] In one embodiment of the present invention, the formula for calculating the conflict degree is: in, express and The degree of conflict and Represents any two valid entities, express and cosine similarity, , ,express and The fused semantic vector ,express and The fused semantic vector express The length of the mold, express The length of the module.

[0012] In one embodiment of the present invention, the formula for calculating the basic score of the effective entity is: in, Indicates the first One valid entity The base score, Indicates the first One valid entity Preset priority weights Indicates the first One valid entity The reliability score.

[0013] A second aspect of this invention provides a standardized intelligent parsing system for home appliance service order information, comprising: The first determining module is used to determine the target basic entity based on the matching degree of each preset basic entity in the order text; wherein, the preset basic entities include: brand entity, category entity and basic specification entity; The second determining module is used to determine the confident entity based on the confidence level of the presence of each preset user demand entity and the target basic entity in the order text; The matching module is used to determine the corresponding target characteristic entity in the preset characteristic entity library based on the confidence entity; The third determining module is used to determine a reliable entity among the confident entities based on the matching degree, the existence confidence degree, and the preset reliability threshold. The first calculation module is used to take the reliability entity and the target characteristic entity as valid entities, and determine the conflict degree based on the cosine similarity between any two valid entities; The second calculation module is used to determine the base score of each valid entity based on the reliability score of the valid entity and the preset priority weight; The third calculation module is used to determine the corrected score of the valid entity based on the base score and the conflict degree when the conflict degree of the valid entity is greater than the preset conflict threshold. The processing module is used to take the valid entities whose base score and modified score are greater than or equal to a preset threshold as the decision result, and output conflict warning and conflict handling information according to the valid entities whose conflict degree is greater than the preset conflict threshold. The generation module is used to generate service prompt tags based on the preset tag template, the valid entity, the decision result, the conflict warning, and the conflict handling information.

[0014] A third aspect of the present invention provides an electronic device, including 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 a standardized intelligent parsing method for home appliance service order information provided in the first aspect of the present invention.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a standardized intelligent parsing method for home appliance service order information provided in the first aspect of the present invention.

[0016] The beneficial effects of this invention are: This invention extracts deterministic entities and identifies ambiguous personalized user needs from unstructured natural language text (such as text notes) input by users. It identifies explicit product-related information and implicit personalized needs, thereby achieving a deep understanding of complex user intentions and ensuring no key service elements are overlooked. The extracted attributes undergo conflict detection and credibility optimization, and core decision attributes are selected accordingly. Finally, a structured tag set is output. This standardized output not only has a unified format, facilitating seamless integration and efficient processing by subsequent systems (such as service scheduling and engineer reminder systems), but also automatically, accurately, and structurally transforms unstructured natural language orders into standard orders. The structured output makes service resource scheduling more precise, and service engineers can intuitively obtain users' core needs and precautions, thus improving the overall operational efficiency and user satisfaction of the skills service platform.

[0017] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 A flowchart illustrating a standardized intelligent parsing method for home appliance service order information provided in an embodiment of the present invention; Figure 2 This is a block diagram of a standardized intelligent parsing system for home appliance service order information provided in an embodiment of the present invention. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0021] like Figure 1 As shown, the first aspect of this invention provides a standardized intelligent parsing method for home appliance service order information, comprising the following steps: Step 11: Determine the target basic entity based on the matching degree of each preset basic entity in the order text.

[0022] The preset basic entities include: brand entities, category entities, and basic specification entities.

[0023] Step 12: Determine the confident entities based on the confidence level of the existence of each preset user demand entity and target basic entity in the order text.

[0024] Step 13: Determine the corresponding target characteristic entity in the preset characteristic entity library based on the confidence entity.

[0025] Step 14: Determine the reliable entity among the target basic entity and the reliable entity based on the matching degree, the existence confidence degree and the preset reliability threshold.

[0026] Step 15: Take the reliability entity and the target characteristic entity as valid entities, and determine the conflict degree based on the cosine similarity between any two valid entities.

[0027] Step 16: Determine the base score for each valid entity based on the reliability score of the valid entity and the preset priority weight.

[0028] Step 17: When the conflict degree of a valid entity is greater than the preset conflict threshold, determine the corrected score of the valid entity based on the base score and conflict degree of the valid entity.

[0029] Step 18: Select the valid entities with base scores and corrected scores that are greater than or equal to a preset threshold as the decision result, and output conflict warning and conflict handling information based on the valid entities with conflict degree greater than the preset conflict threshold.

[0030] Step 19: Generate service prompt labels based on preset label templates, valid entities, decision results, conflict warnings, and conflict handling information.

[0031] In this embodiment, by extracting deterministic entities and identifying ambiguous personalized user needs from unstructured natural language text (such as text notes) input by the user, explicit product-related information and implicit personalized needs are identified, thereby achieving a deep understanding of the user's complex intentions and ensuring that no key service elements are overlooked. Conflict detection and credibility optimization are performed on the extracted attributes, and core decision attributes are selected accordingly. Finally, a structured tag set is output. This standardized output not only has a unified form, facilitating seamless integration and efficient processing by subsequent systems (such as service scheduling and engineer reminder systems), but this invention can automatically, accurately, and structurally convert unstructured natural language orders into standard service prompt tags. The structured output makes service resource scheduling more precise, and service engineers can intuitively obtain users' core needs and precautions through service prompt tags, thereby improving the overall operational efficiency and user satisfaction of the skills service platform.

[0032] Based on the first aspect of the present invention, the second aspect of the present invention provides a more detailed description of a standardized intelligent parsing method for home appliance service order information. The second aspect of the present invention provides a standardized intelligent parsing method for home appliance service order information, applied to a skills service platform, and includes the following steps: Step 21: Determine the target basic entity based on the matching degree of each preset basic entity in the order text.

[0033] The preset basic entities include: brand entities, category entities, and basic specification entities.

[0034] In this step, the preset basic entities are stored in a preset basic thesaurus. The brand entity is the name of the home appliance brand, the category entity is the name of the type of home appliance, and the basic specification entity is the basic operation requirement information for different types of home appliances. For example, the brand entity thesaurus... ={Fotile, Midea, Haier, Gree}, a collection of product category entity terms ={Integrated stove, water heater, air conditioner, refrigerator}, basic specification entity terminology set Relationship with entities in the category entity set, ={[Integrated stove, gas must be turned off first, operation without gas shut off is prohibited, high risk]、[Air conditioner, vacuuming for ≥15 minutes is prohibited, short-term vacuuming is prohibited, medium risk]}.

[0035] Step 21 includes steps 211-212: Step 211: Determine the matching degree of each preset basic entity based on the word frequency of each preset basic entity in the order text and the preset basic weight of each preset basic entity.

[0036] In this step, a rule engine (RE) is used to construct keyword matching rules based on a reverse indexing mechanism. This quickly locates deterministic information (brand, category, basic specification entities) in the order. The accuracy of entity extraction is judged by the matching degree calculation, avoiding incorrect or missed extraction. Matching degree The calculation formula is: in, This indicates the current order text (e.g., "My newly installed Fotile integrated stove in my house has an unpleasant smell when I turn it on," or "My Midea air conditioner isn't cooling, I want to replace it with a Gree"). Indicates a predefined basic entity. express exist In terms of word frequency, keywords in the home appliance context often appear only once. exist If an entity appears in the table, the value is 1; if it does not appear, the value is 0 (as long as the entity appears, there is a basis for matching, regardless of the number of appearances). express The preset base weights are set according to business needs: brand keyword weight 0.7, category keyword weight 0.3, and basic standard keyword weight 0.5, to ensure that the matching degree of different types of attributes is reasonably differentiated. express Belonging to the word set The maximum value of the preset base weights of all entities in the vocabulary (word set) All brands in the list have a weight of 0.7, with a maximum weight of 0.7 (of course, different weights can be set for each brand, and the highest weight will be used); Category keyword set All categories have a weight of 0.3 (with a maximum weight of 0.3) for normalization calculations to ensure that the matching degree is always within the range of [0,1], while ensuring that the matching standards for entities of the same type are consistent.

[0037] Step 212: Determine the target basic entity based on the matching degree and the first threshold. When the matching degree of the preset basic entity is greater than or equal to the first threshold, output the preset basic entity as a target basic entity.

[0038] Match The calculation formula is used to calculate order text. Single preset basic entity The rule matching degree, with a value range of [0,1], can solve the unreasonable problem of "the number of words affecting the matching result" in traditional word set matching, and is more in line with practical application scenarios. Core logic: The matching degree is only related to "whether the entity appears" and "the entity's own weight", and is not related to the number of entities contained in the word set. When the matching degree is ≥0.5, it is considered a successful match, and the corresponding entity can be output, ensuring the accuracy of individual entity extraction (e.g., no matter how many brands are in the brand word set, as long as a certain brand appears in the order, the entity of that brand can be output stably).

[0039] For example, the first threshold is 0.5, and the order text... =“The newly installed Fotile integrated stove in my house has an unpleasant odor when I turn it on”, calculate the matching degree for each: Vocabulary Collection In the group {Fotile, Midea, Haier, Gree}, all brands have a weight of 0.7. , hour =1 (FOTILE appears in the order), substituting into the formula, we get: (≥0.5, matching successful, output a target basic entity) ).

[0040] Vocabulary Collection In the category {integrated stove, water heater, air conditioner, refrigerator}, all categories have a weight of 0.3. , hour =1 (integrated stove appears in the order), substituting into the formula, we get: (≥0.5, matching successful, output target basic entity) ).

[0041] Vocabulary Collection In the categories {[Integrated stove, gas must be shut off first, operation without shutting off gas is prohibited, high risk], and [Air conditioner, vacuuming for ≥15 minutes is prohibited, short-term vacuuming is prohibited, medium risk]}, the weights are both 0.5. , hour =0 (not found in the order) 0×(0.5 / 0.5)=0, so this entity is not output.

[0042] Here, when calculating each type of entity, if other entities have TF=0 and Match=0, then no output is output.

[0043] For example, order text =“My Midea air conditioner isn’t cooling, I want to switch to a Gree,” Brand Keywords ={Fotile, Midea, Haier, Gree} (4 brands, each with a weight of 0.7), calculate the matching degree: : TF=1, matching degree =1×(0.7 / 0.7)=1.0 (≥0.5, output); : TF=1, matching degree =1×(0.7 / 0.7)=1.0 (≥0.5, output); Fotile / Haier : TF=0, matching degree =0 (no output); In this scenario, the order explicitly mentions two brands, and the formula can accurately extract them simultaneously, which aligns with actual business needs (users may mention both the old and new phone brands). This solves the problem of "low matching accuracy due to a large number of keywords" in traditional formulas.

[0044] For example, order text =“My home air conditioner isn’t cooling, please have it repaired as soon as possible”, calculate the matching degree: All brand entities (Fotile, Midea, Haier) have a TF=0, so the matching degree is 0×(0.7 / 0.7)=0. No brand entity is output. The NLP model will further judge based on the context (such as whether there is a historical brand record) to ensure that it matches the actual business logic (the user may not have mentioned the brand).

[0045] The matching degree calculation is completed, and the final output is at least one target basic entity.

[0046] Step 22: Determine the confident entities based on the confidence level of the existence of each preset user demand entity and target basic entity in the order text.

[0047] Step 22 includes steps 221-222: Step 221: Calculate the presence confidence of each preset user demand entity and target basic entity in the order text based on the BERT model.

[0048] In this step, a BERT model fine-tuned with a home appliance service corpus is used to perform entity recognition and intent extraction on ambiguous statements in orders (such as "the elderly are worried about gas accidents at home" and "try to arrive before 3 pm"). Simultaneously, entities not accurately extracted by the rule engine are supplemented and improved (such as ambiguous entities with a rule engine matching score close to 0.5). The model output is the entity confidence score, used to determine whether the extracted entities are valid, as shown in the following formula: in, This indicates the confidence level of the presence of each preset user demand entity and target basic entity in the order text. This represents the order text. express and The fused semantic vector express The model's weight matrix, express The bias term of the model, This represents the normalization function, which transforms the confidence scores of multiple entities output by the model into values ​​within the range of [0,1], and the sum of the confidence scores of all entities is 1, making it easier to intuitively judge the reliability of each entity. The weight matrix and bias terms are obtained after fine-tuning the model using the home appliance service corpus. They are used to optimize the calculation accuracy of semantic vectors and make the model more closely fit the home appliance service scenario.

[0049] Here, for the sake of simplicity and clarity in the formula, both the user requirement entity and the target basic entity are assumed to be represented by letters. This indicates that the preset user demand entity is the preset user demand term set. The entity set includes over 150 frequently used user needs entities (time-related: after 6 PM, weekends; service-related: silent, expedited; scenario-related: elderly at home, new renovations), providing support for extracting implicit and explicit user needs. Of course, to comprehensively extract entities, this step also calculates the basic standardized entity set. The confidence level of the existence of entities in the data.

[0050] Specifically, the open-source pre-trained BERT model is used as the base, and special fine-tuning is done for the home appliance service order scenario. It is trained by feeding a large amount of home appliance order text, user demand language, and attribute tags to make the model adapt to industry terminology and colloquial demand expressions, so as to avoid the problem of inaccurate recognition by general models. It is a fixed-dimensional semantic vector output by the model after semantically fusing the original order text with the entities. Essentially, it transforms "text + entity" into a machine-computable and comparable array of numbers to achieve semantic understanding of requirements, rather than simple keyword matching. This is also the core reason why the model can identify ambiguous requirements.

[0051] here, The specific implementation method is as follows: Step A1, Input Concatenation Processing: Following the standard input format of the BERT model, concatenate the order text... T and entity E The concatenation format is: [CLS] + Order Text T + [SEP] + Entity E + [SEP]. Where [CLS] is the classification tag, used for subsequent semantic aggregation; [SEP] is the separator, used to distinguish order text from entities and avoid semantic confusion.

[0052] Example: Order text T="Fotile integrated cooking range has an odor, my spouse is pregnant staying at home, please come before 10 o'clock", E ="pregnant woman at home", after splicing, the input is: <[BOS_never_used_51bce0c785ca2f68081bfa7d91973934]> "Fotile integrated cooking range has an odor, my spouse is pregnant staying at home, please come before 10 o'clock" "pregnant woman at home" .

[0053] Step A2, word embedding and encoding: the model first splits the spliced input text into individual tokens, then converts each token into an initial word vector, and meanwhile adds position encoding (to distinguish word order) and segment encoding (to distinguish order text and target attribute), converts text information into an initial digital matrix, and retains word order and semantic position information.

[0054] Step A3, Transformer bidirectional semantic encoding: through BERT's multi-layer Transformer encoder, bidirectional semantic understanding is performed, which not only analyzes the context meaning of each token, but also captures the semantic association between the order text and the target attribute, instead of scanning the text in one direction. This step can understand vague expressions: for example, although "my spouse is pregnant staying at home" and "there is a pregnant mother at home" have different keywords, their semantics both point to the entity "pregnant woman at home"; "come as soon as possible" and "come to the door early" both point to the entity "urgent time demand".

[0055] Step A4, semantic vector output: after encoding is completed, the output vector corresponding to the <[BOS_never_used_51bce0c785ca2f68081bfa7d91973934]> marker at the beginning of the input is extracted as the entire "order text T +entity E " fused semantic vector, that is , the vector has a fixed dimension (768 dimensions is commonly used), each number represents a certain semantic feature of the text, and for contents with similar semantics, the corresponding vector distance is closer.

[0056] To put it simply, it is not simply judging "the keywords of the entity E whether appear in the order T ", but understanding "whether the overall meaning of the order T semantically matches the entity E ". For example, if there is no keyword "pregnant woman" in the order, only "my spouse is pregnant waiting for delivery", the model can also identify the corresponding entity through semantic matching; meanwhile, it can distinguish ambiguous expressions and avoid wrong keyword matching.

[0057] The output semantic vector will be input to the subsequent fully connected layer, multiplied by the weight matrix W, added with the bias term b, and then passed through Softmax function for normalization, converted into a confidence score in the 0-1 interval , a higher score indicates a higher semantic matching degree between the order text and the entity.

[0058] Step 222: Determine the confident entities based on the existence confidence level and the second threshold. When the existence confidence level is greater than or equal to the second threshold, the corresponding entity... E For a confident entity .

[0059] For example, the second threshold is 0.8, based on the order text. T =“The Fotile integrated stove we just installed in our new house has an unpleasant odor when it's turned on. My wife is pregnant and we're at home. The technician needs to come and check it before 10 AM today. Please be extremely careful.” (NLP model extracts entities and calculates confidence levels.) First, verify the entities output by the rules engine: =Fotile, =0.93 (≥0.8, valid); =Integrated stove, =0.95 (≥0.8, valid), further confirming the reliability of the extraction results from the rule engine.

[0060] Secondly, extract user demand entities not covered by the rules engine: E =Arrival before 10:00 (expected time) =0.91 (≥0.8, valid); E =Pregnant at home (scenario) =0.88 (≥0.8, valid); E =Safe Operation (Service Requirements) =0.85 (≥0.8, valid); Extracting basic specification entities: E = Cut off the gas supply first (in conjunction with the meanings of "integrated stove" and "safety"). =0.82 (≥0.8, valid), supplementing the canonical entities not extracted by the rule engine.

[0061] After all entity calculations are completed, Entities with a value ≥ 0.8 are used as confidence entities. .

[0062] Step 23: Determine the corresponding target characteristic entity in the preset characteristic entity library based on the confidence entity.

[0063] In this step, the preset characteristic entity library is used to mark the exclusive or special service standards of specific brands in corresponding categories, which are different from the category specifications that are universal for all brands. This includes the "brand-category-exclusive service requirements" triple (e.g., {Samsung, refrigerator, requires special sealant}, {Haier, washing machine, self-cleaning program}). Based on the brand entity and category entity in the confidence entity, the corresponding target characteristic entity is searched in the preset characteristic entity library. A complete triple is considered as a target characteristic entity.

[0064] Step 24: Determine the reliable entity among the reliable entities based on the matching degree, existence confidence degree and preset reliability threshold of the reliable entity.

[0065] In this step, after the rule engine and NLP model extract entities in parallel, a weighted fusion strategy is adopted to combine the advantages of the two types of models (the rule engine is more reliable for deterministic information, and the NLP model is more sensitive to fuzzy semantics) to calculate the final reliability score of the entity. Only when both conditions are met, namely "NLP model confidence ≥ 0.8 (valid entity verification)" and "reliability score ≥ 0.6", can the entity be determined to be valid and temporarily stored.

[0066] Step 24 includes steps 241-242: Step 241: Determine the reliability score for each trusted entity based on its matching degree and existence confidence. Reliability score of a trustworthy entity The calculation formula is: in, This represents the dynamic adjustment coefficient, set according to the entity type: Brand / Category / Basic Specification Entities: α=0.7 (Prioritize rule engines, because these entities have strong determinism and rule matching is more reliable); User Requirement Entities: α=0.3 (Prioritize NLP models, because these entities are mostly fuzzy expressions and semantic understanding is more important).

[0067] The matching degree and confidence degree of the entity can be obtained from the calculations in steps 21 and 22, and the reliability score can then be calculated.

[0068] Example: Brand Entity =Fotile (α=0.7): =0.7×1.0 + 0.3×0.93 = 0.7 + 0.279 =0.979; Category Entities =Integrated stove (α=0.7): =0.7×1.0 + 0.3×0.95 = 0.7 + 0.285 =0.985; User demand entity =Arrival before 10:00 (α=0.3): =0.3×0.0 (The rule engine did not extract this value, so the matching degree is 0) + 0.7×0.91 = 0 + 0.637 = 0.637; User demand entity =Pregnant at home (α=0.3): =0.3×0.0 (The rule engine did not extract this value, so the matching degree is 0) + 0.7×0.88 = 0 + 0.616 = 0.616; Basic Specification Entities = First cut off the gas supply (α=0.7): =0.7×0.0 (The rule engine did not extract it, so the matching degree is 0) + 0.3×0.82 = 0 + 0.246 = 0.246.

[0069] Step 242: Determine the reliability entities based on the reliability score and the preset reliability threshold. The confidence entities whose reliability scores are greater than or equal to the preset reliability threshold are considered as reliability entities.

[0070] For example, the preset reliability threshold is 0.6. ≥0.6、 ≥0.6、 ≥0.6、 If the value is ≥0.6, then the entities Fotile, integrated stove, on-site service before 10:00, and pregnant at home are considered reliable entities.

[0071] The core function of this step is to "filter reliable entities". After all reliable entities are temporarily stored, they enter the subsequent dynamic weight matching and conflict decision-making module. After the decision is completed, they are uniformly mapped to standardized labels and finally output.

[0072] Step 25: Treat the reliability entity and the target characteristic entity as valid entities. Based on any two valid entities Cosine similarity between them to determine the degree of conflict. And conflict types.

[0073] In this step, the conflict degree between any two valid entities is calculated. Generally, when the conflict level is greater than a preset conflict threshold, two entities are determined to be in conflict. Specifically, there are three conflict types: when the conflict level is greater than or equal to the third threshold, the two entities are in significant conflict; when the conflict level is greater than the preset conflict threshold but less than the third threshold, the two entities are in slight conflict; and when the conflict level is less than or equal to the preset conflict threshold, the two entities are not in conflict.

[0074] Define any two valid entities Conflict level between The semantic distance between the two entities is calculated using cosine similarity, which quantifies whether the two entities contradict each other and to what extent they contradict each other. The formula is as follows: in, and Represents any two valid entities, express and Cosine similarity; ,express With valid entities The fused semantic vector ,express With valid entities The fused semantic vector; , All of these can be calculated in step 22; The dot product of two semantic vectors is used to calculate the semantic similarity between two entities. The larger the dot product, the closer the semantics are. express The length of the mold, express The modulus is used to normalize the dot product result, ensuring that the calculated cosine similarity is within the range of [-1, 1].

[0075] This represents the cosine similarity between two semantic vectors, with values ​​ranging from [-1, 1]. The closer to 1, the more consistent the semantics of the two entities; the closer to -1, the more contradictory the semantics. The cosine similarity is converted into the degree of conflict, making the degree of conflict more intuitive (the higher the similarity, the lower the degree of conflict).

[0076] Example, two valid entities =Arrival before 10:00 AM (if explicitly requested by the user) =72-hour calibration period for Fotile integrated cooktop (preset characteristic entity), calculation of conflict degree: Extract the semantic vectors of the two entities: =[0.8, 0.2, 0.1] (core semantic "fast on-site service") =[0.1, 0.8, 0.2] (core semantic "delayed calibration"); Calculate the dot product: ; Calculate the modulus: , ; Calculate cosine similarity: ; Calculate the degree of conflict: , ≥0.5: Significant conflict (two entities have contradictory logic and cannot be satisfied simultaneously, requiring re-coordination); 0.2 < <0.5: Minor conflict (two entities have differences, but can be reconciled and satisfied); ≤0.2: No conflict (both entities have consistent logic and can satisfy the condition simultaneously). Here, This constitutes a significant conflict. A value greater than 0.2 indicates a conflict.

[0077] Step 26: Determine the base score for each valid entity based on the reliability score of the valid entity and the preset priority weight.

[0078] Based on the responsibility logic of home appliance service scenarios, a three-level priority weight is defined (satisfying...) =1), addressing the priority issue of "user needs and industry standards", where priority weights correspond to the importance of various valid entities, as shown in Table 1: base score The calculation formula is: in, Indicates the first i One valid entity Preset priority weights Indicates the first i One valid entity The reliability score is calculated according to step 24.

[0079] Step 27: When the conflict degree of a valid entity is greater than the preset conflict threshold, determine the corrected score of the valid entity based on the base score and conflict degree of the valid entity.

[0080] In this step, it is determined whether the conflict threshold of valid entities is greater than a preset conflict threshold. If it is greater, the base score is adjusted; if it is less than or equal to the preset threshold, no adjustment is needed. (Adjusted Score) The calculation formula is: Wherein, the conflict correction factor = If entity If there is no conflict with other entities (maxC≤0.2), the correction coefficient = 1, and the base score is not corrected; if there is a conflict (maxC>0.2), the correction coefficient <1. ​​The more severe the conflict, the smaller the correction coefficient, and the lower the base score is corrected, so that conflicting entities are given priority.

[0081] Step 28: Select the valid entities with base scores and corrected scores that are greater than or equal to a preset threshold as the decision result, and output conflict warning and conflict handling information based on the valid entities with conflict degree greater than the preset conflict threshold.

[0082] In this step, the decision result means "which entity's service requirement should be prioritized among all currently valid entities" and outputs a conflict warning (if a conflict exists), clarifying the conflict type and handling suggestions to provide clear guidance for service personnel.

[0083] If the base score that does not require correction or the corrected score is greater than or equal to the preset threshold, the service requirements corresponding to the entity are executed directly and the corresponding service process is associated. If the base score that does not require correction or the corrected score is less than 0.4, a "manual review reminder" is triggered to avoid the model misjudging high-risk scenarios. Valid entities with conflicts are searched for their corresponding conflict types and conflict handling information in the preset conflict handling database according to the preset mapping relationship.

[0084] For example, in the preset conflict handling database, time-related conflicts within brand specification conflicts correspond to conflict handling information that coordinates on-site visit times while considering brand specifications. Regular requirement conflicts correspond to conflict handling information that prioritizes security requirements, coordinating regular needs. Multi-user requirement conflicts are sorted by urgency, prioritizing the handling of highly urgent needs. Time-related requirements are categorized as high, and regular requirements as medium. Example: The preset threshold is 0.4. =On-site service before 10:00 (customer's actual needs) =0.6, Final=0.85); =Fotile integrated cooktops require airtightness testing (characteristic entity, =0.3, Final=0.85); =Pregnant women need maternal and infant safety services at home (user demand entity) =0.6, Final=0.82). =Fotile integrated cooktop 72-hour calibration period (characteristic entity, =0.3, Final=0.90), based on the calculations in the steps above. and The degree of conflict is There is a significant conflict.

[0085] Calculate the base score for each entity (accurate and error-free): ; ; ; ; calculate and Corrected score: : =0.623 (and (Conflict), score after correction =0.51×(1-0.623)=0.51×0.377≈0.192 (<0.4, triggering manual review); : =0.623 (and (Conflict), score after correction =0.27×(1-0.623)=0.27×0.377≈0.102 (<0.4, triggering manual review); Decision result: Select the base score (no adjustment required) and the adjusted score (scores greater than or equal to 0.4). (0.492) As the final decision, namely, to implement "pregnant women needing maternal and infant safety services at home"; no conflict. If the score is less than 0.4, a manual review is triggered to determine whether the service corresponding to the entity needs to be executed.

[0086] Preferably, the priority of the valid entity corresponding to the maximum score among those greater than or equal to 0.4 is determined as the highest priority. In the example above, The highest priority is given to "Maternal and Infant Safety Services for Pregnant Women at Home".

[0087] Result determination: but , , All scores were <0.4, therefore a conflict warning was issued: "Conflict type: User's on-site visit before 10:00 conflicts with the brand's 72-hour calibration period; Conflict handling information: Coordinate on-site visit time while taking into account brand specifications; Manual review result: Fotile integrated stove needs to be tested for air tightness."

[0088] Example of a conflict-free scenario: If the order does not contain (72-hour calibration period), then , , No conflicts, no corrections required: Score =0.51 (≥0.4) =0.255 (<0.4) =0.492 (≥0.4), the decision result is to execute "on-site visit before 10:00 + maternal and infant safety service + airtightness test of Fotile integrated stove (manual verification result)", with no conflict warning. Among them, The highest base score has the highest priority, and the "on-site visit before 10:00" option will be executed first.

[0089] Step 29: Generate service prompt labels based on preset label templates, valid entities, decision results, conflict warnings, and conflict handling information.

[0090] In this step, valid entities, decision results, conflict warnings, and conflict handling information are matched to preset label templates to form the final service prompt labels.

[0091] For example, the preset label template is as follows: curly braces represent variable positions, corresponding to the results extracted / calculated by the above methods.

[0092] Brand: {Brand Entity} | Category: {Category Entity} | Product Model: {Order-Associated Model} Key operational points: {Brand / Category Basic Specifications} | Safety requirements: {General category prohibitions} Expected Time: {User Time Entity} | Service Requirements: {User Scenario / Requirement Entity} | Priority: {Automatically Determined to be Highest} Conflict Warning Type: {Conflict Type} | Conflict Handling: {Preset Script} | Risk Level: {High / Medium / Low} Each entity carries its own type identifier, indicating whether it belongs to a brand entity or a category entity. Order information can also extract product information and match it to tags. Security requirements can be automatically matched with relevant security operating procedures based on brand and category. Risk levels in alerts are judged and matched based on the degree of conflict: C ≥ 0.5 is high risk, 0.2 < C < 0.5 is medium risk, and C ≤ 0.2 is no risk.

[0093] Example, original order text: Fotile integrated stove has an odor, pregnant woman is at home, please come before 10 am, please be careful.

[0094] Valid entity: Brand = Fotile, Category = Integrated stove, User needs = On-site service before 10:00 AM / Pregnant woman at home, Basic specifications = Gas shut off for integrated stoves first, Brand category characteristics = Air tightness test / 72-hour calibration period.

[0095] Generate service notification labels: Brand: Fotile | Category: Integrated Cooktop | Product Model: JZT-Y2T Operating Instructions: Airtightness Test | Safety Requirements: Do not operate while the gas supply is still running. Expected arrival time: Before 10:00 AM | Service requirements: Maternal and infant safety services | Priority: High Conflict warning type: User's on-site visit before 10:00 AM conflicts with the brand's 72-hour calibration period | Conflict handling: Coordinate on-site visit time while adhering to brand regulations | Risk level: High.

[0096] In this embodiment, unstructured natural language orders can be automatically, accurately, and structurally converted into standard service prompt labels. The structured output makes service resource scheduling more precise, and service engineers can intuitively obtain users' core needs and precautions, thereby improving the overall operational efficiency and user satisfaction of the skills service platform.

[0097] like Figure 2 As shown, a third aspect of the present invention provides a standardized intelligent parsing system for home appliance service order information, comprising: The first determining module 31 is used to determine the target basic entity based on the matching degree of each preset basic entity in the order text; wherein, the preset basic entities include: brand entity, category entity and basic specification entity; The second determining module 32 is used to determine the confident entity based on the confidence level of the existence of each preset user demand entity and target basic entity in the order text; Matching module 33 is used to determine the corresponding target characteristic entity in the preset characteristic entity library based on the confidence entity; The third determining module 34 is used to determine the reliable entity among the reliable entities based on the matching degree, the existence confidence degree and the preset reliability threshold; The first calculation module 35 is used to take the reliability entity and the target characteristic entity as valid entities and determine the conflict degree based on the cosine similarity between any two valid entities. The second calculation module 36 is used to determine the basic score of each valid entity based on the reliability score of the valid entity and the preset priority weight; The third calculation module 37 is used to determine the corrected score of a valid entity based on its base score and conflict degree when the conflict degree of a valid entity is greater than a preset conflict threshold. The processing module 38 is used to take valid entities with base scores and corrected scores greater than or equal to preset thresholds as decision results, and output conflict warnings and conflict handling information based on valid entities with conflict degrees greater than preset conflict thresholds. The generation module 39 is used to generate service prompt labels based on preset label templates, valid entities, decision results, conflict warnings, and conflict handling information.

[0098] In one embodiment of the present invention, determining the target basic entity based on the matching degree of each preset basic entity in the order text includes: The matching degree of each preset basic entity is determined based on the word frequency of each preset basic entity in the order text and the preset basic weight of each preset basic entity; The target basic entity is determined based on the matching degree and the first threshold.

[0099] In one embodiment of the present invention, determining a confident entity based on the confidence level of the presence of each preset user demand entity and target basic entity in the order text includes: The presence confidence of each preset user demand entity and target basic entity in the order text is calculated based on the BERT model. Confidential entities are determined based on the existence confidence level and a second threshold.

[0100] In one embodiment of the present invention, determining a reliable entity from the reliable entities based on the matching degree, the existence confidence degree, and a preset reliability threshold includes: The reliability score for each confident entity is determined based on the matching degree and the existence confidence of the confident entity. Reliability entities are determined based on reliability scores and preset reliability thresholds.

[0101] In one embodiment of the present invention, the formula for calculating the confidence level is as follows: in, This indicates the confidence level of the presence of each preset user demand entity and target basic entity in the order text. This represents the order text. This represents the preset user requirement entity and the target basic entity. express and The fused semantic vector express The model's weight matrix, express The bias term of the model, This represents the normalization function.

[0102] In one embodiment of the present invention, the formula for calculating the degree of conflict is: in, express and The degree of conflict and Represents any two valid entities, express and cosine similarity, , ,express and The fused semantic vector ,express and The fused semantic vector express The length of the mold, express The length of the module.

[0103] In one embodiment of the present invention, the formula for calculating the basic score of a valid entity is: in, Indicates the first One valid entity The base score, Indicates the first One valid entity Preset priority weights Indicates the first One valid entity The reliability score.

[0104] A fourth aspect of the present invention provides an electronic device, including 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 above-described intelligent parsing method for standardized home appliance service order information provided by the present invention.

[0105] The fifth aspect of this invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described intelligent parsing method for standardized home appliance service order information provided in this invention.

[0106] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage system located remotely from the aforementioned processor.

[0107] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware systems.

[0108] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0109] For system / electronic device embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be found in the description of the method embodiments.

[0110] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A standardized intelligent parsing method for home appliance service order information, characterized in that, Includes the following steps: The target basic entity is determined based on the matching degree of each preset basic entity in the order text; wherein, the preset basic entities include: brand entity, category entity and basic specification entity; Based on the confidence level of the presence of each preset user demand entity and the target basic entity in the order text, a confidence entity is determined; wherein, the preset user demand entity is an entity in the preset user demand term set; The corresponding target characteristic entity is determined in the preset characteristic entity library based on the confidence entity; the preset characteristic entity library is used to mark the exclusive or special service standards of a specific brand in the corresponding category, which are different from the category specifications that are common to all brands. The corresponding target characteristic entity is found in the preset characteristic entity library based on the brand entity and category entity in the confidence entity. A complete triple is a target characteristic entity. Based on the matching degree and the existence confidence degree of the trusted entity, a reliability score is determined for each trusted entity; wherein, the reliability score of the trusted entity... The calculation formula is: in, This represents a dynamic adjustment coefficient, set according to the entity type: Brand, Category, and Basic Specification entities: α = 0.7; User Requirement entities: α = 0.

3. Indicates a confident entity. This represents the order text. Indicates the degree of matching. This indicates the presence of a confidence level; Based on the reliability score and the preset reliability threshold, a reliable entity is determined; The reliability entity and the target characteristic entity are taken as valid entities, and the conflict degree is determined based on the cosine similarity between any two valid entities. The base score for each valid entity is determined based on the reliability score of the valid entity and the preset priority weight; When the conflict degree of the valid entity is greater than the preset conflict threshold, the corrected score of the valid entity is determined based on the base score and conflict degree of the valid entity. The effective entities whose base score and modified score are greater than or equal to a preset threshold are used as the decision results, and conflict warning and conflict handling information are output according to the effective entities whose conflict degree is greater than the preset conflict threshold. Service prompt tags are generated based on the preset tag template, the valid entities, the decision results, the conflict warning, and the conflict handling information; The formula for calculating the degree of conflict is: in, express and The degree of conflict and Represents any two valid entities. express and cosine similarity, , , indicating order text and The fused semantic vector , indicating order text and The fused semantic vector express The length of the mold, express The length of the module.

2. The method as described in claim 1, characterized in that, The step of determining the target basic entity based on the matching degree of each preset basic entity in the order text includes: The matching degree of each preset basic entity is determined based on the word frequency of each preset basic entity in the order text and the preset basic weight of each preset basic entity; The target basic entity is determined based on the matching degree and the first threshold.

3. The method as described in claim 1, characterized in that, The step of determining the confidence entity based on the presence confidence of each preset user demand entity and the target basic entity in the order text includes: The presence confidence of each preset user demand entity and the target basic entity in the order text is calculated based on the BERT model. Confidential entities are determined based on the aforementioned confidence level and the second threshold.

4. The method as described in claim 3, characterized in that, The formula for calculating the existence confidence level is as follows: in, This indicates the confidence level of the presence of each preset user demand entity and the target basic entity in the order text. This represents the order text. This represents the preset user requirement entity and the target basic entity. express and The fused semantic vector express The model's weight matrix, express The bias term of the model, This represents the normalization function.

5. The method as described in claim 1, characterized in that, The formula for calculating the basic score of the effective entity is as follows: in, Indicates the first One valid entity The base score, Indicates the first One valid entity Preset priority weights Indicates the first One valid entity The reliability score.

6. A standardized intelligent analysis system for home appliance service order information, characterized in that, To perform the method of claim 1, comprising: The first determining module is used to determine the target basic entity based on the matching degree of each preset basic entity in the order text; wherein, the preset basic entities include: brand entity, category entity and basic specification entity; The second determining module is used to determine the confident entity based on the confidence level of the presence of each preset user demand entity and the target basic entity in the order text; The matching module is used to determine the corresponding target characteristic entity in the preset characteristic entity library based on the confidence entity; The third determining module is used to determine a reliable entity among the confident entities based on the matching degree, the existence confidence degree, and the preset reliability threshold. The first calculation module is used to take the reliability entity and the target characteristic entity as valid entities, and determine the conflict degree based on the cosine similarity between any two valid entities; The second calculation module is used to determine the base score of each valid entity based on the reliability score of the valid entity and the preset priority weight; The third calculation module is used to determine the corrected score of the valid entity based on the base score and the conflict degree when the conflict degree of the valid entity is greater than the preset conflict threshold. The processing module is used to take the valid entities whose base score and modified score are greater than or equal to a preset threshold as the decision result, and output conflict warning and conflict handling information according to the valid entities whose conflict degree is greater than the preset conflict threshold. The generation module is used to generate service prompt tags based on the preset tag template, the valid entity, the decision result, the conflict warning, and the conflict handling information.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the standardized intelligent parsing method for home appliance service order information as described in any one of claims 1 to 5.

8. A 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 standardized intelligent parsing method for home appliance service order information as described in any one of claims 1 to 5.

Images