A user demand analysis and quantification method and system based on concept metaphor analysis

Abstract

This invention discloses a user demand analysis and quantification method and system based on conceptual metaphor parsing, belonging to the field of natural language processing. It establishes a product attribute classification system to semantically segment user comment corpora, and then builds a conceptual metaphor analysis framework based on the classification system and semantic segmentation units. This framework performs a two-stage parsing process involving conceptual metaphor identification and structured parsing, yielding metaphor parsing results. Quantification is then performed to obtain a metaphor set including source domain labels, target domain labels, metaphor type, sentiment polarity score, expectation intensity, and demand type. The metaphor set is aggregated and statistically analyzed from a product dimension, achieving demand analysis and quantification. This method comprehensively covers metaphorical expressions in user comments and, through a systematic mapping and quantification mechanism from metaphor structure to user demand type and demand intensity, generates multi-dimensional, comparable attribute-level structured demand analysis results from large-scale user comments.

Description

Technical Field

[0001] This invention relates to the field of natural language processing, specifically to a method and system for user demand analysis and quantification based on conceptual metaphor parsing. Background Technology

[0002] User reviews are a crucial source of information for product design and experience optimization. With the development of e-commerce and social media platforms, users generate a large amount of textual reviews about product experiences on various platforms. How to systematically extract user needs information from these unstructured texts has become an important research topic at the intersection of natural language processing and product design.

[0003] Demand analysis methods based on user reviews mainly revolve around sentiment analysis. Generally, existing methods typically include: analyzing product attributes, identifying review statements related to those attributes and determining their sentiment polarity, and then generating a result summary. For example, patent CN116226332A discloses a metaphor generation method and system based on conceptual metaphor theory, including: collecting text to construct a metaphor generation corpus, including pairs of metaphorical and non-metaphorical sentences; obtaining the semantic primitives of metaphorical words and substitutes, respectively serving as the source and target domains; constructing a source-target domain mapping table; using each non-metaphorical sentence as a basis, combining the source-target domain mapping table to construct the input of a metaphor model, using the corresponding metaphorical sentence as the output, training and obtaining the metaphor model; obtaining the semantic primitives of the verbs in the non-metaphorical sentences to obtain the target domain; based on the source-target domain mapping table, obtaining all mappings for the target domain, and sampling a target-source domain mapping relationship according to probability labels; constructing the input of the metaphor model based on the non-metaphorical sentence and the target-source domain mapping relationship, inputting it into the metaphor model, and generating verbal metaphorical sentences.

[0004] Building upon metaphor generation, analyzing and quantifying the generated metaphors can more effectively help understand user needs and enable targeted product design. Early research primarily involved automatically extracting attributes from large-scale comment corpora on specific topics based on statistical information such as word frequency, and then classifying each attribute based on a sentiment dictionary. In recent years, neural networks have also received widespread attention in attribute-level sentiment analysis due to their strong representation learning capabilities. Furthermore, existing technologies have also seen patent solutions related to text structuring and implicit semantic processing. For example, the invention patent with publication number CN118133844A discloses an evaluation method and apparatus for the implicit semantic recognition capability of a large language model, including: constructing a test dataset of implicit semantics of comment text, which contains comment text and labels on multiple dimensions; using a large language model to predict the implicit semantics, implicit intent, and logical consistency of the comment text in the test dataset; using a semantic encoder to semantically encode the prediction results and labels, and then evaluating the model's semantic understanding capability based on the encoding results; using Cohen's Kappa statistical analysis algorithm to compare the model's implicit intent prediction results with the labels and evaluate the model's implicit intent recognition capability; using the F1 score calculation method to evaluate the model's recognition capability in logical consistency; and combining the three capabilities to obtain the model's comprehensive implicit semantic recognition capability, promoting the application and development of the field of comment text analysis, and focusing on the evaluation of the implicit semantic recognition capability of large language models.

[0005] However, the aforementioned existing technologies still have the following shortcomings in the in-depth needs analysis of user comments: First, existing attribute-level sentiment analysis methods struggle to effectively handle metaphorical expressions in reviews. Users often describe their product experience using non-literal language such as "a feeling of being pushed back," "a feeling of being suspended," or "like being on a boat," implying that their needs are often hidden in the cross-conceptual mapping between the source and target domains. Existing methods primarily focus on attribute extraction and sentiment polarity classification, lacking the ability to identify and analyze metaphorical expressions, which can easily lead to the omission of deeper needs information.

[0006] Second, many existing metaphor detection methods mainly output metaphor or non-metaphorical judgment results, without further parsing the metaphorical expression into structured data containing elements such as source domain, target domain, and mapping type, making it difficult to connect with subsequent requirements analysis and quantitative processing.

[0007] Third, existing methods lack a systematic mapping and quantification mechanism from metaphorical structures to user need types and intensity. User needs for products cover different levels such as functionality, experience, and symbolism, while existing sentiment analysis methods usually only output sentiment polarity, making it difficult to provide multi-dimensional, comparable, and structured insights into needs, and thus difficult to directly support product design decisions. Summary of the Invention

[0008] To address the shortcomings of existing attribute-level sentiment analysis methods, which fail to identify and parse metaphorical expressions in user comments, resulting in the omission of deep-seated needs information implicit in cross-conceptual domain mapping relationships, and the inability of existing metaphor detection methods to convert metaphorical expressions into structured data containing elements such as source domain, target domain, and mapping type, making it impossible to connect metaphor recognition results with subsequent needs analysis processing, and the lack of a systematic mapping and quantification mechanism from metaphorical structure to user needs type and intensity, thus failing to generate multi-dimensional and comparable attribute-level structured needs analysis results from large-scale user comments, this invention provides a user needs analysis and quantification method and system based on conceptual metaphor parsing. This method can fully cover metaphorical expressions in user comments and generate multi-dimensional and comparable attribute-level structured needs analysis results from large-scale user comments through a systematic mapping and quantification mechanism from metaphorical structure to user needs type and intensity.

[0009] To achieve the above-mentioned objectives, an embodiment provides a user needs analysis and quantification method based on conceptual metaphor parsing, comprising the following steps: The original user comment corpus is preprocessed, a product attribute classification system is established, and the preprocessed comment corpus is semantically segmented based on the product attribute classification system to obtain the smallest semantic analysis unit. Based on the product attribute classification system and the smallest semantic analysis unit, a conceptual metaphor analysis framework is established, which includes target domain labels, source domain labels, metaphor type labels, and metaphor structured representation rules. This framework is used to perform a two-stage analysis on the semantic analysis unit, which includes conceptual metaphor identification and structured parsing, to obtain metaphor analysis results. Based on the metaphor analysis results, a complete set of metaphors is obtained through quantitative processing, including the definition of the demand classification system, the mapping of demand types, and the calculation of expectation intensity. By aggregating and statistically analyzing the product attribute dimensions of the complete set of metaphors, the intensity of demand is comprehensively quantified, and structured user demand insights are generated, thus completing the analysis and quantification of user needs.

[0010] Preferably, the preprocessing includes: removing garbled characters, emoticons, and marker characters without actual semantic content based on the user's original comment corpus; Delete comments whose character length is less than a preset character length threshold and do not contain a valid product experience description; By combining rule matching with large language model discrimination, we can identify and remove advertising, promotional material, repetitive or templated expressions, and comments that are irrelevant to the product experience. Calculate the text similarity between comments, identify comments with a text similarity higher than a preset text similarity threshold as near-duplicate comments, and retain the comment with the most complete and valid product experience description in each group of near-duplicate comments.

[0011] Preferably, the establishment of the product attribute classification system includes: based on the semantic content of the preprocessed comment corpus, identifying the product attribute expressions involved in the comment corpus, and obtaining product attribute candidate words; The candidate words for product attributes are standardized, synonymized, and hierarchically organized to form a product attribute classification system; the product attribute classification system is used to describe the product attribute dimensions corresponding to the user review corpus.

[0012] Preferably, semantic segmentation is based on the following rules: For comment corpora that cannot be segmented, the entire comment corpus is treated as the smallest semantic analysis unit; If the segmented result does not meet the preset structured format requirements, semantic segmentation will be re-executed; If the results of multiple semantic segmentations still do not meet the preset structured format requirements, the comment corpus will be removed.

[0013] Through the semantic segmentation operation described above, each subsequent comment corpus to be analyzed has a clear attribute affiliation, thereby avoiding the ambiguity in metaphor recognition caused by multiple attributes mixed in the same comment corpus and significantly improving the accuracy of analysis.

[0014] Preferably, the product attribute classification system is used as the target domain label to point to the product attribute category; Typical metaphorical expressions appearing in the smallest semantic analysis unit are summarized and normalized and merged in combination with domain knowledge to obtain source domain labels; Metaphorical type tags are used to describe the mapping between the source domain and the target domain; Based on target domain labels, source domain labels, and metaphor type labels, a structured model is used to create metaphor structured representation rules, which are used to convert metaphorical natural language expressions in preprocessed comment corpora into structured data objects.

[0015] By pre-defining the aforementioned label space, problems such as label drift, inconsistent semantic classification, or multiple names for the same expression can be avoided in the subsequent parsing process of the conceptual metaphor analysis framework, thereby improving the consistency, comparability, and quantifiability of the subsequent identification results.

[0016] More preferably, the source domain label includes: force and matter, used to represent an expression that compares the product's performance to physical force, material properties, or tactile characteristics; Objects and behaviors are used to represent expressions that draw analogies between the state characteristics, action patterns, or response tendencies of objects and the performance of products. Natural phenomena are used to describe product performance by analogy to natural states such as wind, water, flow, and impact. Space and container are used to represent product expressions by analogy to spatial relationships of enclosing, compressing, opening, and constricting; Direction and movement are used to express product performance by analogy to the relationship of direction or movement, such as up and down, front and back, light and heavy, following, and stability. Interaction and perception are used to describe expressions that compare product performance to interactive relationships, feedback methods, degree of coordination, or perceptual impressions.

[0017] More preferably, the metaphor type label includes structural metaphor, directional metaphor, and ontological metaphor; The structural metaphor uses experiential structures to systematically analogize product attributes; The directional metaphor refers to the use of spatial orientation, weight, height, and movement trends to express evaluation tendencies. The ontological metaphor refers to the materialization, behavioralization, or objectification of abstract performance states, interactive experiences, or subjective feelings.

[0018] By adopting the above metaphor types and limiting them to a preset set, metaphorical expressions in different commentary corpora can have a unified classification standard, which facilitates subsequent structured parsing with consistent rules.

[0019] Preferably, performing concept metaphor recognition includes: Using the smallest semantic analysis unit as input, combined with the established conceptual metaphor analysis framework, the large language model is invoked to traverse the content words in the comment corpus, retaining words with substantial semantic carrying capacity, and obtaining the filtered content words; Based on each filtered content word, the contextual meaning and the regular literal meaning of the content word in the current context are given, and it is determined whether there is a cross-conceptual domain analogy between the contextual meaning and the regular literal meaning. If the analogy exists, the content word is marked as a metaphor candidate trigger word, and a set of metaphor candidates is obtained for subsequent metaphor structured parsing.

[0020] Preferably, performing structured parsing includes: Using the contextual semantics of the smallest semantic analysis unit containing the metaphor candidate trigger word in the metaphor candidate set as input, and the conceptual metaphor analysis framework as constraint, the large language model is guided to make attribution judgments within a preset category range, and output structured results.

[0021] Furthermore, the structured results include: source domain labels, target domain labels, metaphor type labels, sentiment polarity scores, overall parsing confidence, and implicit expectation semantic descriptions.

[0022] Preferably, the definition of the demand classification system includes: functional demands, experiential demands, and symbolic demands; The functional requirements refer to the requirements put forward by users regarding the objective performance, operational results, or physical indicators of the product. The experiential needs refer to the subjective perceptions and emotional experiences that users develop during the usage process. The symbolic needs refer to the user's needs related to identity recognition, brand tone, social expression, or aesthetic meaning formed around the product.

[0023] Preferably, the requirement type mapping includes: based on the requirement classification system definition, using a multi-factor joint mapping method, calculating the requirement matching score on the metaphor analysis results to achieve requirement type mapping.

[0024] Furthermore, the formula for calculating the demand matching score is as follows: , in, Score the matching score to the demand. To label a given source domain Target domain tags Metaphorical type tags Under the given conditions, the metaphor belongs to the demand type. The conditional probability, Represents implicit expectation semantic description With target domain tags Next Demand Type semantic prototype template Normalized semantic similarity between them; Indicates target domain label Demand type under conditions The prior probability, , , These are the weighting coefficients. Represents the label index variable. This is an index variable for the demand type.

[0025] Preferably, the expected intensity calculation includes: comprehensively considering three factors—language intensity marker, absolute value of sentiment polarity, and metaphor novelty—and using the Sigmoid function to normalize the linear combination result to the (0,1) interval. The calculation formula is as follows: , in, For the desired strength, Score the language intensity. Represents the absolute value of emotional polarity. The score indicates the novelty of the metaphor. , , These are non-negative weighting coefficients; This is an adjustable offset constant.

[0026] Preferably, the aggregation and statistics of the complete metaphor set along the product attribute dimension includes: Based on the complete set of metaphors, a subset of metaphors that satisfy any attribute in the product attribute classification system is selected from the target domain labels; The demand attention, demand polarity distribution, comprehensive demand intensity, demand type distribution, and source domain interpretation distribution of the corresponding attributes of the metaphor subset are calculated as indicators, and a structured user demand insight result is constructed based on the indicators.

[0027] Preferably, the demand attention level is used to measure the significance of user discussions about a certain product attribute in the comment corpus, and the demand attention level of attribute a is calculated by combining frequency ratio and inverse comment coverage. : , in, This indicates the total number of comments in the corpus. This indicates that at least one target domain is any attribute. The number of comments in the smallest semantic analysis unit. Reflecting any attribute The relative proportion of all metaphorical needs expressed; The demand polarity distribution is used to describe the evaluation direction structure of users for a certain product attribute. The evaluation direction is divided into positive, negative and neutral according to a preset threshold. The demand polarity distribution is represented by a triple based on the proportion of demand expression in each evaluation direction. The comprehensive demand intensity mentioned above is used to measure the overall urgency of demand for a certain product attribute in user reviews. Specifically, it is obtained by weighted fusion of normalized demand attention, mean expectation intensity, and polarity concentration. The aforementioned demand type distribution is used to reveal the nature of demand carried by different product attributes. This is achieved by establishing a distribution matrix with demand type as rows and product attribute as columns. The matrix elements in this distribution matrix represent the demand nature of each attribute. The normalized proportion of each demand type is used to characterize the distribution of demand types. The aforementioned source domain interpretation distribution is used to aggregate attribute dimensions based on source domain labels in the complete metaphor set, establishing an interpretation distribution matrix with source domain categories as rows and product attributes as columns. The matrix elements in the interpretation distribution matrix represent the interpretation distribution of any attribute. The normalized percentage of source domain categories is used to characterize the conceptual framework used by users when describing different product attributes.

[0028] Furthermore, the structured user needs insight results may take the form of a structured table, matrix, or sorted list.

[0029] On the other hand, the present invention also provides a user needs analysis and quantification system based on conceptual metaphor parsing. This system uses the aforementioned user needs analysis and quantification method based on conceptual metaphor parsing, and includes: The data preprocessing module is used to preprocess the original user comment corpus, establish a product attribute classification system, and perform semantic segmentation on the preprocessed comment corpus based on the product attribute classification system to obtain the smallest semantic analysis unit. The metaphor analysis module is used to establish a conceptual metaphor analysis framework based on the product attribute classification system and the smallest semantic analysis unit, which includes target domain labels, source domain labels, metaphor type labels and metaphor structured representation rules. It is used to perform a two-stage analysis on the semantic analysis unit, which includes conceptual metaphor identification and structured parsing, to obtain the metaphor analysis results. The demand quantification module is used to obtain a complete set of metaphors based on the metaphor analysis results through quantification processing, including the definition of the demand classification system, the mapping of demand types, and the calculation of expected intensity. The aggregation analysis module is used to perform aggregate statistics on the product attribute dimensions of the complete metaphor set, realize the comprehensive quantification of demand intensity, and complete the analysis and quantification of user needs. The storage and output module is used to store comment data, attribute classification system, structured parsing results, complete metaphor set and aggregation analysis results, and generate structured user demand insight results.

[0030] Compared with the prior art, the beneficial effects of the present invention include at least the following: (1) The invention achieves the identification and structured parsing of metaphorical expressions in user comments. Through a two-stage identification process, the invention first locates candidate metaphorical trigger words at the word level, and then completes structured attribution judgment within a preset tag space at the sentence level. This helps reduce the identification bias problem caused by directly judging metaphors in the entire sentence. By pre-establishing a unified tag constraint system, the consistency and standardization of metaphorical expression parsing in large-scale comments are improved.

[0031] (2) A six-tuple structured representation of metaphors was proposed, and a quantitative analysis mechanism from metaphor structure to user needs was established. Each conceptual metaphor is uniformly represented by a six-tuple (source domain label, target domain label, metaphor type, sentiment polarity score, expectation intensity, and need type), and the need type determination and expectation intensity calculation are completed through multi-factor joint mapping, so that the need information implied in natural language can be transformed into statistically quantifiable, aggregateable, and comparable structured data.

[0032] (3) It can output multi-dimensional attribute-level structured requirements analysis results. This invention can output results such as requirements attention, requirements polarity distribution, comprehensive requirements intensity, requirements type distribution and source domain distribution. Compared with existing methods that only provide emotional polarity statistics, the output information is richer and helps to provide a basis for product design optimization and experience improvement. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0034] Figure 1 This is a flowchart illustrating the user needs analysis and quantification method based on conceptual metaphor parsing provided in the embodiments.

[0035] Figure 2 A detailed flowchart illustrating the user needs analysis and quantification method based on conceptual metaphor parsing provided for this embodiment.

[0036] Figure 3 The diagram illustrates the structure and process of a complete set of metaphors represented by a six-tuple, as provided in the embodiment.

[0037] Figure 4 This is a schematic diagram of the structure of a user needs analysis and quantification system based on conceptual metaphor parsing, provided for an embodiment. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of this invention.

[0039] The concept of this invention lies in addressing the shortcomings of existing attribute-level sentiment analysis methods, which fail to identify and parse metaphorical expressions in user comments, resulting in the omission of deep-seated needs information implicit in cross-conceptual domain mapping relationships. Existing metaphor detection methods can only determine whether text constitutes a metaphorical usage, lacking the ability to convert metaphorical expressions into structured data containing elements such as source domain, target domain, and mapping type, making it impossible to connect metaphor identification results with subsequent needs analysis processing. Furthermore, existing methods lack a systematic mapping and quantification mechanism from metaphorical structure to user needs type and intensity, failing to generate multi-dimensional, comparable attribute-level structured needs analysis results from large-scale user comments. This invention provides a user demand analysis and quantification method and system based on conceptual metaphor parsing. It establishes a product attribute classification system to semantically segment user comment corpora, and then builds a conceptual metaphor analysis framework based on the classification system and semantic segmentation units. This framework performs a two-stage parsing process involving conceptual metaphor identification and structured parsing to obtain metaphor parsing results. Quantification is then performed to obtain a metaphor set including source domain labels, target domain labels, metaphor type, sentiment polarity score, expectation intensity, and demand type. The metaphor set is aggregated and statistically analyzed from a product dimension, achieving demand analysis and quantification. This method can comprehensively cover metaphorical expressions in user comments and, through a systematic mapping and quantification mechanism from metaphor structure to user demand type and demand intensity, generates multi-dimensional, comparable attribute-level structured demand analysis results from large-scale user comments.

[0040] In the embodiments, such as Figure 1 As shown, a user needs analysis and quantification method based on conceptual metaphor analysis is provided, including the following steps: S1. Preprocess the original user comment corpus, establish a product attribute classification system, and perform semantic segmentation on the preprocessed comment corpus based on the product attribute classification system to obtain the smallest semantic analysis unit. This step is used to clean and standardize the original user comment corpus, and on this basis, establish a product attribute classification system to further segment the comments into the smallest semantic analysis units with a single attribute affiliation. Details are as follows: like Figure 2 As shown, the input is the original set of user comments. The output is a collection of high-quality comments. Product Attribute Classification System With the corresponding set of minimal semantic analysis units ,in, A subset of user comments For user comment index variables, A subset of high-quality comments For high-quality comment indexing variables, For product attribute classification system index, As an index variable for the product attribute classification system, A subset of the smallest semantic analysis units, This is the index variable for the smallest semantic analysis unit.

[0041] Specifically, firstly, the original set of user comments... Preprocessing is performed. This preprocessing includes: removing garbled characters, emoticons, and marker characters without actual semantic content; deleting comments with a character length lower than a preset character length threshold and lacking a valid product experience description; identifying and removing comments that are advertisements, promotional material, repetitive or templated expressions, or irrelevant to the product experience using a combination of rule matching and large language model discrimination; calculating the text similarity between comments, identifying comments with a text similarity higher than a preset text similarity threshold as near-duplicate comments, and retaining the comment with the most complete valid product experience description from each group of near-duplicate comments. After the above processing, a high-quality comment set is obtained. .

[0042] In obtaining a high-quality comment collection Then, a product attribute classification system is constructed based on the aforementioned comment set. The specific method is: based on sets. The semantic content and contextual expression of the comment text are analyzed to identify candidate product attribute words involved in the comments. These candidate words are then standardized, synonymized, and hierarchically organized to form an attribute classification system oriented towards the target product domain. This attribute classification system describes the product attribute dimensions corresponding to the user comments. In automotive application scenarios, the attribute classification system may include attribute categories such as power / acceleration, chassis / handling, braking, interior / materials, NVH (Noise, Vibration and Harshness) / quietness, intelligence / interaction, space / layout, and appearance / design. To ensure consistency in subsequent processing, the target domain labels in subsequent steps are all derived from this attribute classification system. Selected from the options.

[0043] In constructing an attribute classification system Then, a collection of high-quality comments. Each comment in the process undergoes semantic segmentation to extract the minimum semantic analysis unit (MAU). The MAU is defined as a comment fragment that can independently express a relatively complete semantic meaning and corresponds to only one product attribute category. For each comment... Using large language models in attribute classification systems Under the constraints of [unclear], clause segmentation, clause merging, and attribute attribution judgment are performed, outputting several minimal semantic analysis units. Let [unclear] be the [unclear]... The smallest semantic analysis unit is ,in Indicates the first The text content of the smallest semantic analysis unit, This represents the product attribute tag corresponding to the smallest semantic analysis unit. All smallest semantic analysis units constitute a set. If a comment cannot be reliably segmented, it is treated as a single minimal semantic analysis unit. If the output of the large language model does not meet the preset structured format requirements, the segmentation is re-executed. If a valid structured result still cannot be obtained, the comment is removed from the subsequent metaphor recognition steps. This removal only affects the set of minimal semantic analysis units. The composition does not change the high-quality comment set already output in step S1. And its counting. This step ensures that each subsequent text to be analyzed has a clear attribute attribution, thereby avoiding ambiguity in metaphor identification caused by multiple attributes mixed together in the same comment.

[0044] S2. Based on the product attribute classification system and the smallest semantic analysis unit, establish a conceptual metaphor analysis framework that includes target domain labels, source domain labels, metaphor type labels, and metaphor structured representation rules. This framework is used to perform a two-stage parsing process, including conceptual metaphor identification and structured parsing, on the semantic analysis unit to obtain the metaphor parsing results.

[0045] like Figure 2 As shown, this step pre-establishes a constrained conceptual metaphor analysis framework, which serves as a unified criterion for subsequent metaphor recognition and structured parsing by the large language model. The conceptual metaphor analysis framework includes at least: target domain labels, source domain labels, metaphor type labels, and metaphor structured data representation rules. By pre-defining the aforementioned label space, problems such as label drift, inconsistent semantic classification, or multiple names for similar expressions during the generation process of the large language model can be avoided, thereby improving the consistency, comparability, and quantifiability of subsequent recognition results.

[0046] The target domain represents the product attribute category to which the metaphor points. This step directly reuses the product attribute classification system established in S1 as the target domain label, meaning that each metaphor identified subsequently must belong to a certain attribute category in the product attribute classification system.

[0047] The source domain represents the conceptual domain from which users use analogies to describe product attributes. This step, based on the set of minimal semantic analysis units output by S1, summarizes the typical metaphorical expressions that appear within it, and combines domain experience to standardize and merge the source domains, forming preset source domain labels. In a preferred embodiment, the source domain labels include the following six categories: force and matter, object and behavior, natural phenomena, space and container, direction and motion, and interaction and perception.

[0048] Among them, "Force and Matter" is used to represent expressions that use physical force, material properties, or tactile characteristics to analogize product performance; "Object and Behavior" is used to represent expressions that use the state characteristics, action patterns, or response tendencies of an object to analogize product performance; "Natural Phenomena" is used to represent expressions that use natural states such as wind, water, flow, and impact to analogize product performance; "Space and Container" is used to represent expressions that use spatial relationships such as envelopment, compression, openness, and contraction to analogize product performance; "Direction and Movement" is used to represent expressions that use directional or movement relationships such as up and down, front and back, light and heavy, following, and stability to analogize product performance; and "Interaction and Perception" is used to represent expressions that use interactive relationships, feedback methods, coordination levels, or perceptual impressions to analogize product performance. The source domains identified in subsequent steps must all be selected from the aforementioned source domain label set.

[0049] After defining the target and source domain labels, metaphor type labels are further established. These metaphor types describe the mapping between the source and target domains. In this method, metaphor types are limited to three categories: structural metaphors, directional metaphors, and ontological metaphors. Structural metaphors refer to using a relatively concrete experiential structure to systematically analogize product attributes; directional metaphors refer to using spatial orientation, weight, height, and movement trends to express evaluation tendencies; and ontological metaphors refer to materializing, behaving, or objectifying abstract performance states, interactive experiences, or subjective feelings. By limiting metaphor types to a preset set, metaphorical expressions in different review texts can have a unified classification standard, facilitating subsequent structured parsing with consistent rules.

[0050] Based on the target domain label, source domain label, and metaphor type label mentioned above, this step defines the core data structure of this method, namely the metaphor six-tuple. For the first... The identified conceptual metaphors are defined as follows: ;in, Indicates the first A metaphorical source domain tag, Indicates the first A metaphorical target domain label, Indicates the first The metaphor type of a metaphor, Indicates the first The metaphorical emotional polarity score, Indicates the first The intensity of expectation of a metaphor, Indicates the first The metaphorical demand type is defined as follows. In the six-tuple, the source domain label, target domain label, metaphor type, and sentiment polarity are directly output by the subsequent two-stage parsing, while the expectation intensity and demand type are filled in by step S3.

[0051] In addition, an implicit expectation semantic description field can be added outside the six-tuple to serve as an auxiliary semantic basis when calculating expectation strength and demand mapping, but this field does not participate in the main structure definition as a core necessary component. By adopting a unified six-tuple representation, the metaphorical natural language expressions in the original comments can be converted into statistically quantifiable, aggregateable, and comparable structured data objects, providing a unified input for the demand quantification and result output in steps S3 and S4.

[0052] In the embodiment, the set of minimum semantic analysis units output in step S1 Perform conceptual metaphor identification and structured parsing. The input is a set of minimal semantic analysis units with attribute labels. The output includes the target domain labels, source domain labels, metaphor type labels, and metaphor structured representation rules pre-established in step S2; the output is a set of metaphor parsing results. .

[0053] For each smallest semantic analysis unit The process employs a two-stage identification approach: first, word-level metaphor candidate identification is performed, followed by sentence-level conceptual metaphor structure parsing, thereby converting metaphorical expressions in natural language into structured results that can be quantified for subsequent processing. The purpose of this two-stage process is to avoid the identification drift problem caused by directly judging metaphors as a whole sentence, allowing the model to first locate potential metaphor trigger words and then complete structured attribution judgments within a constrained label space.

[0054] Specifically, firstly, for each smallest semantic analysis unit Perform word-level metaphor candidate identification. Call a large language model to analyze the text. The system iterates through the content words, excluding function words and retaining only nouns, verbs, adjectives, and adverbs—words with substantial semantic capacity. For each content word, it provides its contextual meaning and its literal meaning in the current context, and determines whether there is a cross-conceptual domain analogy between them. If such a mapping exists, the word is marked as a metaphor candidate trigger word. The output of this stage is a set of candidate options, where each candidate option includes at least the target word, its contextual meaning, its literal meaning, and a marker indicating whether it constitutes a metaphorical usage. If no metaphor candidate option exists in a minimum semantic analysis unit, the unit is determined not to contain an identifiable conceptual metaphor and will not proceed to subsequent metaphor structure parsing.

[0055] After word-level candidate identification is completed, sentence-level conceptual metaphor structure parsing is performed on the smallest semantic analysis unit containing at least one metaphor candidate. Specifically, the candidate trigger word and its context semantics within the smallest semantic analysis unit are used as input, while the source domain label, target domain label, and metaphor type label established in step S2 are used as constraints and input into the large language model. The large language model is required to complete the attribution judgment within a preset category range and output structured results.

[0056] like Figure 3 As shown, the structured result includes at least the following fields: source domain label Target domain tags Metaphorical type Emotional polarity score and overall analytical confidence level It can also include an output field that implicitly describes the expected semantics. This is used to characterize the deep user expectations implied by the metaphorical expression, and is invoked in step S3 when calculating the intensity of expectation. The source domain label... The metaphor type is determined by the conceptual category to which the literal meaning of the candidate trigger word belongs, and its value is only allowed to be selected from the source domain labels preset in step S2; Only three types of metaphors are allowed: structural metaphors, directional metaphors, and ontological metaphors, which are preset in step S2.

[0057] The emotional polarity score This value characterizes the direction and intensity of a user's evaluation of the target domain, ranging from a continuous interval [-1, 1]. Positive values ​​represent positive evaluations, negative values ​​represent negative evaluations, and zero values ​​represent neutrality or no clear evaluation tendency. A larger absolute value indicates higher evaluation intensity. Its calculation comprehensively considers the following semantic signals: evaluative vocabulary, the sentiment valence of the source domain containing candidate trigger words, and the consistency between implicit expectations and the current description. The sentiment polarity score, along with the source domain label, target domain label, and metaphor type, is output synchronously in the same statement-level parsing to ensure semantic consistency among the fields.

[0058] To improve the stability of the structured parsing results, this step employs a consistency sampling strategy to estimate the confidence level of each parsing result and combines this with a target domain conflict resolution mechanism to determine the final output. Specifically, for the same smallest semantic analysis unit... Independently call large language models under the same constraints Second-rate( ),get Group-structured output. Statistics. The number of times the source domain label, target domain label, and metaphor type are all consistent in the output. The initial confidence level is defined as: .

[0059] In obtaining After the first output, the final values ​​of the source domain label, target domain label, and metaphor type are determined by majority voting. Then, it is checked whether the target domain label obtained by majority voting matches the attribute label assigned in step S1. If consistent, then adopt the result directly, and the final confidence level is... If they are inconsistent, a constrained re-parsing will be triggered, meaning the target domain will be explicitly specified in the prompt. Then call the model independently. Next, recount the number of times the three items match and update. If the target domain obtained by majority voting after re-parsing is the same as... If they are still inconsistent, then retain the attribute labels assigned in step S1. As the target domain label for this metaphor, the final confidence level is also lowered to: , in, This is a preset discount factor used to reflect the negative impact of target domain conflicts on the reliability of parsing.

[0060] Set a preset confidence threshold When the final confidence level of a certain parsing result When the result is obtained, it is retained and output to the subsequent step S3; when If such a result is not found, it will be discarded and will not be included in subsequent requirements quantification processing. By combining the above-mentioned consistency sampling with target domain conflict handling, this step can still automatically filter out low-stability parsing results without relying on manual review as a necessary step.

[0061] After the above two-stage processing, for each identified conceptual metaphor, a corresponding structured parsing result is generated and used to fill the four components of the metaphor six-tuple: source domain label, target domain label, metaphor type, and sentiment polarity score. If an implicit expected semantic description field is output, it is stored as an auxiliary field along with the corresponding six-tuple. All structured parsing results constitute a set. This serves as the input for step S3, which performs the demand type mapping and expected intensity calculation.

[0062] S3. Based on the metaphor analysis results, a complete set of metaphors is obtained through quantitative processing, including the definition of the demand classification system, the mapping of demand types, and the calculation of expectation intensity.

[0063] In the embodiments, such as Figure 2 As shown, this step is used to further quantify the metaphor structure parsing result output in step S2, completing the remaining two components in the metaphor six-tuple, namely the demand type. and expected intensity The input is the set of structured parsing results output from step S2, which is then filled in to generate a complete set of metaphorical six-tuples for attribute-level requirement aggregation and statistical analysis in step S4. The output is a complete set of metaphorical six-tuples. Each six-tuple .

[0064] This step includes three parts: defining the demand classification system, mapping demand types, and calculating the expected intensity.

[0065] First, a demand classification system for this method is established. Specifically, user needs are divided into three categories: functional needs, experiential needs, and symbolic needs. Functional needs refer to user requirements regarding the product's objective performance, operational results, or physical indicators, such as power response, braking effect, and space utilization. Experiential needs refer to the subjective perceptions and emotional experiences formed by users during use, such as comfort, a sense of security, enjoyment of operation, and smooth interaction. Symbolic needs refer to user needs related to identity recognition, brand tone, social expression, or aesthetic meaning formed around the product. These three categories of needs together constitute a demand type tag set. Each subsequent metaphor must be mapped to a requirement type in the aforementioned tag set.

[0066] After defining the requirement classification system, requirement type mapping is performed on each metaphor analysis result output in step S2. This method employs a multi-factor joint mapping approach. Specifically, for any metaphor analysis result... For each candidate requirement type Calculate its demand matching score Specifically, when step S2 outputs the implicit expected semantic description field... In this case, it is used as an auxiliary input in the demand matching calculation: , in, This indicates that, given the source domain label, target domain label, and metaphor type, the metaphor belongs to the requirement type. The conditional probability; Represents implicit expectation semantic description With the target domain Next Demand Type semantic prototype template Normalized semantic similarity between them; Indicates in the target domain Demand type under conditions The prior probability; , , Let be the weighting coefficient, satisfying ;when When unavailable, make and redistribute the weights to and .

[0067] The semantic prototype template A pre-built set of text descriptions used to characterize the target domain. Next Demand Type The typical expected semantics. The prototype semantic representation is initialized by typical textual descriptions written by domain experts, and obtained through induction, aggregation, or iterative updating combined with implicit expected descriptions of corresponding demand types in the annotated corpus. Semantic similarity. Cosine similarity calculation based on text embedding representation.

[0068] The conditional probability and prior probability can be estimated offline based on pre-established labeled samples, or they can be initialized based on expert experience during the initial deployment of the system and then updated offline as the corpus accumulates; the semantic prototype template This metaphor describes the typical semantic features of different demand types within a specific target domain. For this metaphor, the demand type is determined by the category with the highest demand matching score, i.e.: .

[0069] After completing the demand type mapping, the expected strength of the metaphor is further calculated. The expected intensity is used to measure the urgency or intensity of a user's need reflected in the metaphorical expression, and its value ranges from (0,1). In this method, the expected intensity comprehensively considers three factors: language intensity markers, absolute value of sentiment polarity, and metaphorical novelty, and uses the Sigmoid function to normalize the linear combination result to the (0,1) interval: , in, Represents the Sigmoid function; Indicates the score of language intensity markers; This represents the sentiment polarity score output in step S2. The absolute value; The score indicates the novelty of the metaphor. , , These are non-negative weighting coefficients; It is an adjustable offset constant used to control the desired intensity baseline, so that the desired intensity can approach zero when all input factors are low.

[0070] Among them, language strength marker score This is used to characterize the presence of degree adverbs, emphatic structures, or repetitive modifiers in the text that reinforce the user's intended expression. In this embodiment, the text, the smallest semantic analysis unit, is weighted based on a preset intensity dictionary and syntactic rules to obtain a normalized intensity marker score. If the text contains reinforcing expressions such as "very," "obvious," "especially," "very," or "simply," and these expressions apply to descriptions relevant to the target domain, a higher score is assigned; otherwise, if no significant intensity marker is found, then... When needed, combine the implicit expected semantic description field output from step S2. Semantic verification is performed to determine whether intensity words truly point to the target attribute requirements, in order to avoid misscoring irrelevant rhetoric.

[0071] Absolute value of emotional polarity The continuous score of sentiment polarity output from step S2 The absolute value is obtained. Regardless of whether the evaluation is positive or negative, as long as the user's expression in the target domain has a high emotional intensity, it indicates that the user has a clearer and stronger focus or expectation on that attribute, and therefore can be obtained through... It reflects the prominence of the expressed demand.

[0072] Metaphorical Novelty Score This is used to measure the relative rarity of the current metaphorical expression in the entire corpus. Preferably, novelty is calculated at the level of the "source domain-target domain-metaphor type" triad. Let... Represents a ternary combination The novelty score of a metaphor can be defined as the frequency of its occurrence in the entire corpus. , in, To prevent extremely small positive numbers with a denominator of zero, this embodiment may also use a fixed target domain. Under certain conditions, intra-domain normalization is performed to reduce the impact of the imbalance in the number of comments with different attributes. The higher the novelty score, the less common the metaphor is among similar comments, and it can be used as one of the candidate signals that users use unconventional expressions to describe their perceptual state. This indicator, together with language intensity markers and sentiment polarity intensity, participates in the expectation intensity estimation, but is not considered as a sufficient condition for unmet needs on its own.

[0073] After the above calculations, the demand type can be obtained for each metaphor parsing result. and expected intensity This forms a complete metaphorical six-tuple: .

[0074] The complete set of metaphorical six-tuples This serves as the input for step S4, which involves statistical analysis of product attribute-level demand attention, polarity distribution, overall demand intensity, and demand type distribution.

[0075] S4. Perform aggregated statistics on the product attribute dimensions of the complete metaphor set to achieve comprehensive quantification of demand intensity, generate structured user demand insight results, and complete user demand analysis and quantification.

[0076] In this embodiment, this step is used to aggregate and statistically analyze the complete set of metaphorical six-tuples output in step S3 based on product attribute dimensions, generating structured user demand insights for product design, experience optimization, and demand decision-making. The input is the complete set of metaphorical six-tuples. Each six-tuple The output includes the demand attention, demand polarity distribution, comprehensive demand intensity, demand type distribution matrix, and source domain interpretation distribution matrix for each product attribute.

[0077] Specifically, regarding the product attribute classification system any attribute in First, filter out the target domain tags that meet the requirements. The complete set of metaphorical six-tuples constitutes the subset corresponding to this attribute: ; Let its base be . The total number of complete metaphorical six-tuples is Based on this, the demand attention, demand polarity distribution, comprehensive demand intensity, demand type distribution, and source domain interpretation distribution of this attribute are calculated respectively.

[0078] The demand attention level is used to measure the significance of user discussion about a product attribute in the comment corpus. Considering that using only frequency as an indicator can easily lead to overweighting of generic attributes that are "mentioned in almost all comments," this method uses a combination of frequency ratio and inverse comment coverage to calculate the attribute. demand attention : , in, This represents the total number of comments in the high-quality comment set output in step S1. This indicates that at least one target domain is included as an attribute. The number of comments in the smallest semantic analysis unit; where the first term Reflecting attributes In the relative proportion of all metaphorical needs expressed, the second item is used to appropriately reduce the weight of common attributes that are too broad in scope, thereby highlighting the attributes that are centrally discussed among the user group. Preferably, all attributes can be... Further max-min normalization is performed to obtain the normalized demand focus. This is used for subsequent comprehensive demand intensity calculations.

[0079] The demand polarity distribution describes the directional structure of users' evaluations of a particular product attribute. Let the threshold for determining emotional polarity be... ,in This is a preset constant; when When, the metaphor is judged as a positive evaluation; when When, it is judged as a negative evaluation; when At that time, it was judged as a neutral evaluation. Based on this, the attribute was... The polarity distribution of demand is represented by triples: , in, , , , And satisfy .

[0080] The triples represent user-defined attributes. The proportion of positive, negative, and neutral demand expressions can directly reflect the satisfaction structure and degree of controversy of this attribute.

[0081] The comprehensive demand intensity is used to measure the overall urgency of demand for a product attribute in user reviews. Considering that a single indicator cannot simultaneously reflect the degree of attention, the intensity of expression, and the clarity of evaluation direction, this method weights and fuses three quantities: normalized demand attention, the mean of attribute expected intensity, and polarity concentration, to obtain the attribute... Comprehensive demand intensity : , in, , , The weighting coefficients are non-negative and satisfy the following conditions: ; For attributes The normalization requirement needs attention. For attributes The mean expected strength of all metaphorical six-tuples is defined as: ; For attributes The effective directional concentration, used to characterize whether positive and negative evaluations of this attribute are significantly concentrated in a certain direction when non-neutral evaluations exist, is defined as: , in, To prevent extremely small positive numbers with a denominator of zero, as shown in the above formula, when all metaphors under a certain attribute are neutral evaluations, ,thereby This indicates that the attribute lacks effective directional information; when an attribute has a high proportion of non-neutral ratings, and the positive and negative ratings are clearly concentrated in one direction, A value close to 1 indicates a relatively clear demand direction for that attribute; when the ratio of positive to negative evaluations is close... A value close to 0 indicates that users have significant disagreements about this attribute.

[0082] Through the , and Joint modeling can improve the comprehensive demand intensity. It simultaneously reflects the discussion's intensity, emotional level, and directional certainty regarding the attribute. Preferably, it can be done according to... Sort all attributes in descending order to generate an attribute-level requirement priority list.

[0083] To further reveal the nature of demand embodied by different product attributes, this step establishes a demand type-product attribute distribution matrix for the complete metaphorical six-tuple set. The matrix is ​​structured with demand type as the row dimension and product attribute as the column dimension, with matrix elements... Indicates in the attribute Above, demand type The normalized proportion is defined as: , in, And for any attribute All satisfy: .

[0084] This matrix can characterize the main conceptual frameworks that users use when describing different product attributes. For example, if an attribute mainly corresponds to the "force and matter" source domain, it means that users tend to understand the attribute using frameworks such as physical strength, material texture, or mechanical support. If an attribute mainly corresponds to the "object and behavior" or "interaction and perception" source domain, it means that users are more inclined to describe the product performance corresponding to the attribute as an object with action, response, or interactive feedback characteristics.

[0085] Furthermore, attribute-dimensional aggregation is performed on the source domain tags in the complete metaphor six-tuple set to establish a source domain-product attribute interpretation distribution matrix. The matrix is ​​structured with source domain categories as the row dimension and product attributes as the column dimension, with matrix elements... Indicates in the attribute Above, source domain category The normalized proportion is defined as: , in, This represents the defined source domain category, and for any attribute... All satisfy: .

[0086] This matrix can characterize the main conceptual frameworks that users use when describing different product attributes. For example, if an attribute mainly corresponds to the "force and matter" source domain, it means that users tend to understand the attribute using frameworks such as physical strength, material texture, or mechanical support. If an attribute mainly corresponds to the "object and behavior" or "interaction and perception" source domain, it means that users are more inclined to describe the product performance corresponding to the attribute as an object with action, response, or interactive feedback characteristics.

[0087] In this embodiment, attributes can be further calculated. In a certain source domain Average expected intensity : ; This is used to identify which conceptual framework corresponds to a stronger expression of demand for a particular product attribute. The results can help designers understand the deeper experiential needs related to the attribute from the perspective of user cognitive schemas.

[0088] After completing the above calculations, this step outputs a structured result set consisting of the aforementioned indicators for each attribute. In this embodiment, the output results of all attributes can be organized into a structured table, matrix, or sorted list to support product design optimization, configuration adjustment, marketing positioning, or experience improvement decisions. Among them, demand attention, demand polarity distribution, comprehensive demand intensity, and demand type distribution belong to the main decision output, used to answer "which attribute deserves priority, what is its demand direction, and what type of demand mainly drives it"; source domain distribution belongs to the interpretation layer output, used to answer "what conceptual framework users mainly use to understand this attribute".

[0089] Based on the same technical concept, this application also provides a user demand analysis and quantification system based on conceptual metaphor parsing. The system can be deployed on a server, cloud platform, local computing device, or other electronic device. It is used to automatically process, parsed, mapped, and output results from user comment corpora using the aforementioned user demand analysis and quantification method based on conceptual metaphor parsing. The system includes: The data preprocessing module receives raw user comment corpora and cleans, denoises, deduplicates, and filters invalid comments to obtain a high-quality comment set. Simultaneously, it performs semantic segmentation on the comments, extracts the smallest semantic analysis units, and assigns corresponding attribute labels to each smallest semantic analysis unit in conjunction with the product attribute classification system. The metaphor analysis module, based on a pre-defined conceptual metaphor analysis framework, performs a two-stage analysis on the smallest semantic analysis unit, including conceptual metaphor identification and structured parsing, to identify metaphorical expressions in comments and output corresponding structured parsing results; the structured parsing results include at least source domain labels, target domain labels, metaphor type, and sentiment polarity score; The demand quantification module is used to obtain a complete set of metaphors based on the metaphor parsing results through quantification processing including the definition of the demand classification system, demand type mapping, and expectation intensity calculation. The complete set of metaphors is in the form of a six-tuple and includes at least the source domain label, target domain label, metaphor type, sentiment polarity score, expectation intensity, and demand type. The aggregation analysis module is used to perform aggregate statistics on the product attribute dimensions of the complete metaphor set, realize the comprehensive quantification of demand intensity, and complete the analysis and quantification of user needs; the structured user demand insight results include the demand attention, demand polarity distribution, demand intensity, and demand type distribution results corresponding to each attribute. The storage and output module stores comment data, attribute classification systems, structured parsing results, complete metaphor sets, and aggregated analysis results, and generates structured user demand insights in the form of tables, sorted lists, matrices, or visualizations.

[0090] Each module can be implemented by one or more processors calling program instructions in memory; each module can also be merged, split, or deployed in a distributed manner according to actual deployment needs.

[0091] In summary, this invention provides a user demand analysis and quantification method and system based on conceptual metaphor parsing. It establishes a product attribute classification system to semantically segment user comment corpora, and then builds a conceptual metaphor analysis framework based on the classification system and semantic segmentation units. This framework performs a two-stage parsing process involving conceptual metaphor identification and structured parsing to obtain metaphor parsing results. Quantification is then performed to obtain a metaphor set including source domain labels, target domain labels, metaphor type, sentiment polarity score, expectation intensity, and demand type. The metaphor set is aggregated and statistically analyzed from a product dimension, achieving demand analysis and quantification. This method can comprehensively cover metaphorical expressions in user comments and, through a systematic mapping and quantification mechanism from metaphor structure to user demand type and demand intensity, generates multi-dimensional, comparable attribute-level structured demand analysis results from large-scale user comments.

[0092] The specific embodiments described above illustrate the technical solution and beneficial effects of the present invention in detail. It should be understood that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for user needs analysis and quantification based on conceptual metaphor parsing, characterized in that, Includes the following steps: The original user comment corpus is preprocessed, a product attribute classification system is established, and the preprocessed comment corpus is semantically segmented based on the product attribute classification system to obtain the smallest semantic analysis unit. Based on the product attribute classification system and the smallest semantic analysis unit, a conceptual metaphor analysis framework is established, which includes target domain labels, source domain labels, metaphor type labels, and metaphor structured representation rules. This framework is used to perform a two-stage analysis on the semantic analysis unit, which includes conceptual metaphor identification and structured parsing, to obtain metaphor analysis results. Based on the metaphor analysis results, a complete set of metaphors is obtained through quantitative processing, including the definition of the demand classification system, the mapping of demand types, and the calculation of expectation intensity. By aggregating and statistically analyzing the product attribute dimensions of the complete set of metaphors, the intensity of demand is comprehensively quantified, and structured user demand insights are generated, thus completing the analysis and quantification of user needs.

2. The user needs analysis and quantification method based on conceptual metaphor parsing according to claim 1, characterized in that, The establishment of the product attribute classification system includes: based on the semantic content of the preprocessed comment corpus, identifying the product attribute expressions involved in the comment corpus, and obtaining product attribute candidate words; The candidate words for product attributes are standardized, synonymized, and hierarchically organized to form a product attribute classification system; the product attribute classification system is used to describe the product attribute dimensions corresponding to the user review corpus.

3. The user needs analysis and quantification method based on conceptual metaphor parsing according to claim 1, characterized in that, The product attribute classification system is used as the target domain label to point to the product attribute category; Typical metaphorical expressions appearing in the smallest semantic analysis unit are summarized and normalized and merged in combination with domain knowledge to obtain source domain labels; Metaphor type tags are used to describe the mapping between the source domain and the target domain, including structural metaphor, directional metaphor, and ontological metaphor. The structural metaphor uses experiential structures to make systematic analogies to product attributes. The directional metaphor refers to expressing evaluation tendencies using spatial orientation, weight, height, and movement trends. The ontological metaphor refers to materializing, behaving, or objectifying abstract performance states, interactive experiences, or subjective feelings. Based on target domain labels, source domain labels, and metaphor type labels, a structured model is used to create metaphor structured representation rules, which are used to convert metaphorical natural language expressions in preprocessed comment corpora into structured data objects.

4. The user needs analysis and quantification method based on conceptual metaphor parsing according to claim 1, characterized in that, Performing conceptual metaphor identification includes: Using the smallest semantic analysis unit as input, combined with the established conceptual metaphor analysis framework, the large language model is invoked to traverse the content words in the comment corpus, retaining words with substantial semantic carrying capacity, and obtaining the filtered content words; Based on each filtered content word, the contextual meaning and the regular literal meaning of the content word in the current context are given, and it is determined whether there is a cross-conceptual domain analogy between the contextual meaning and the regular literal meaning. If the analogy exists, the content word is marked as a metaphor candidate trigger word, and a set of metaphor candidates is obtained for subsequent metaphor structured parsing.

5. The user needs analysis and quantification method based on conceptual metaphor parsing according to claim 4, characterized in that, Performing structured parsing includes: Using the contextual semantics of the smallest semantic analysis unit containing the metaphor candidate trigger word in the metaphor candidate set as input, and the conceptual metaphor analysis framework as constraint, the large language model is guided to make attribution judgments within a preset category range, and output structured results.

6. The user needs analysis and quantification method based on conceptual metaphor parsing according to claim 1, characterized in that, The aforementioned demand type mapping includes: based on the demand classification system definition, using a multi-factor joint mapping method, calculating the demand matching score based on the metaphor analysis results to achieve demand type mapping; The formula for calculating the demand matching score is as follows: , Score the matching score to the demand. To label a given source domain Target domain tags Metaphorical type tags Under the given conditions, the metaphor belongs to the demand type. The conditional probability, Represents implicit expectation semantic description With target domain tags Next Demand Type semantic prototype template Normalized semantic similarity between them; Indicates target domain label Demand type under conditions The prior probability, , , These are the weighting coefficients. Represents the label index variable. This is an index variable for the demand type.

7. The user needs analysis and quantification method based on conceptual metaphor parsing according to claim 1, characterized in that, The calculation of expected intensity includes: comprehensively considering three factors: language intensity markers, absolute value of sentiment polarity, and metaphor novelty, and using the Sigmoid function to normalize the linear combination result to the (0,1) interval. The calculation formula is as follows: ， in, For the desired strength, Score the language intensity. Represents the absolute value of emotional polarity. The score indicates the novelty of the metaphor. , , These are non-negative weighting coefficients; This is an adjustable offset constant.

8. The user needs analysis and quantification method based on conceptual metaphor parsing according to claim 1, characterized in that, The aforementioned aggregation and statistical analysis of the complete set of metaphors along the product attribute dimension includes: Based on the complete set of metaphors, a subset of metaphors that satisfy any attribute in the product attribute classification system is selected from the target domain labels; The demand attention, demand polarity distribution, comprehensive demand intensity, demand type distribution, and source domain interpretation distribution of the corresponding attributes of the metaphor subset are calculated as indicators, and a structured user demand insight result is constructed based on the indicators.

9. The user needs analysis and quantification method based on conceptual metaphor parsing according to claim 8, characterized in that, The aforementioned demand attention level is used to measure the significance of user discussion about a certain product attribute in the comment corpus. It is calculated by combining frequency ratio and inverse comment coverage. demand attention : ， in, This indicates the total number of comments in the corpus. This indicates that at least one target domain is any attribute. The number of comments in the smallest semantic analysis unit. Reflecting any attribute The relative proportion of all metaphorical needs expressed; The demand polarity distribution is used to describe the evaluation direction structure of users for a certain product attribute. The evaluation direction is divided into positive, negative and neutral according to a preset threshold. The demand polarity distribution is represented by a triple based on the proportion of demand expression in each evaluation direction. The comprehensive demand intensity mentioned above is used to measure the overall urgency of demand for a certain product attribute in user reviews. Specifically, it is obtained by weighted fusion of normalized demand attention, mean expectation intensity, and polarity concentration. The aforementioned demand type distribution is used to reveal the nature of demand carried by different product attributes. This is achieved by establishing a distribution matrix with demand type as rows and product attribute as columns. The matrix elements in this distribution matrix represent the demand nature of each attribute. The normalized proportion of each demand type is used to characterize the distribution of demand types. The aforementioned source domain interpretation distribution is used to aggregate attribute dimensions based on source domain labels in the complete metaphor set, establishing an interpretation distribution matrix with source domain categories as rows and product attributes as columns. The matrix elements in the interpretation distribution matrix represent the interpretation distribution of any attribute. The normalized percentage of source domain categories is used to characterize the conceptual framework used by users when describing different product attributes.

10. A user needs analysis and quantification system based on conceptual metaphor parsing, characterized in that, The user needs analysis and quantification system uses the user needs analysis and quantification method based on conceptual metaphor parsing as described in any one of claims 1-9, including: The data preprocessing module is used to preprocess the original user comment corpus, establish a product attribute classification system, and perform semantic segmentation on the preprocessed comment corpus based on the product attribute classification system to obtain the smallest semantic analysis unit. The metaphor analysis module is used to establish a conceptual metaphor analysis framework based on the product attribute classification system and the smallest semantic analysis unit, which includes target domain labels, source domain labels, metaphor type labels and metaphor structured representation rules. It is used to perform a two-stage analysis on the semantic analysis unit, which includes conceptual metaphor identification and structured parsing, to obtain the metaphor analysis results. The demand quantification module is used to obtain a complete set of metaphors based on the metaphor analysis results through quantification processing, including the definition of the demand classification system, the mapping of demand types, and the calculation of expected intensity. The aggregation analysis module is used to perform aggregate statistics on the product attribute dimensions of the complete metaphor set, realize the comprehensive quantification of demand intensity, and complete the analysis and quantification of user needs. The storage and output module is used to store comment data, attribute classification system, structured parsing results, complete metaphor set and aggregation analysis results, and generate structured user demand insight results.

Images