Intelligent Assignment System for Requirement Priorities Based on Requirement Semantic Analysis

Abstract

The present application provides an intelligent demand priority assignment system based on requirement semantic parsing, which relates to the field of intelligent requirement management. It uses artificial intelligence-based natural language analysis and coding technologies to perform semantic parsing on a first requirement text and a second requirement text. At the same time, business rules are extracted and semantic parsing is performed on them. Then, a requirement difference semantic parsing feature between the semantic parsing features of the first requirement text and the second requirement text is calculated. Based on this, the priorities of the first requirement and the second requirement are intelligently assigned according to the business rule-guided bidirectional attention requirement difference semantic representation between the requirement difference semantic parsing feature and the business rule semantic parsing feature. In this way, the accuracy and rationality of demand priority evaluation can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of intelligent requirement management, and more specifically, to an intelligent requirement priority assignment system based on requirement semantic parsing. Background Art

[0002] Today, with the rapid development of digitalization, enterprises and organizations are faced with a vast amount of requirements, covering aspects such as software development, business process optimization, and product iteration. Efficient requirement management is the key to ensuring project success and sustainable business development. By determining requirement priorities, the allocation and assignment of limited resources can be optimized to ensure that the most important requirements are processed in a timely manner, thereby enhancing the project success rate and resource utilization efficiency.

[0003] However, traditional methods for determining requirement priorities, such as judgment based on experience, simple voting, or ranking according to customer importance, etc., are no longer able to meet the modern complex and ever-changing business requirements. These methods often lack scientificity and objectivity and are easily interfered by subjective factors, resulting in inaccurate judgment of requirement priorities, which in turn affects the project schedule, cost, and quality. In addition, modern business rules are often complex and variable, and traditional technologies face many challenges in extracting and applying these rules, making it difficult to effectively combine business rules with requirements, thus unable to ensure that the priority assignment conforms to the actual business requirements.

[0004] Therefore, there is a need for an intelligent requirement priority assignment system based on requirement semantic parsing. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide an intelligent requirement priority assignment system based on requirement semantic parsing.

[0006] According to one aspect of the present application, a demand priority intelligent assignment system based on demand semantic parsing is provided, which includes: a demand text acquisition module for acquiring a first demand text and a second demand text; a demand text semantic parsing module for performing semantic parsing on the first demand text and the second demand text to obtain a first demand text semantic parsing encoded vector and a second demand text semantic parsing encoded vector; a business rule semantic parsing module for extracting business rules and performing semantic parsing on the business rules to obtain a business rule semantic parsing encoded vector; a demand difference calculation module for calculating a demand difference semantic parsing encoded vector between the first demand text semantic parsing encoded vector and the second demand text semantic parsing encoded vector; a demand difference two-way evaluation module for performing a business rule-based demand difference two-way attention evaluation on the demand difference semantic parsing encoded vector and the business rule semantic parsing encoded vector to obtain a demand difference evaluation semantic encoded vector under the business rule; and an assignment result generation module for obtaining an assignment result based on the demand difference evaluation semantic encoded vector under the business rule.

[0007] Compared with the prior art, the demand priority intelligent assignment system based on demand semantic parsing provided by the present application uses artificial intelligence-based natural language analysis and encoding technologies to perform semantic parsing on the first demand text and the second demand text. At the same time, business rules are extracted and semantic parsing is performed on them. Then, the demand difference semantic parsing features between the first demand text semantic parsing features and the second demand text semantic parsing features are calculated, and based on this, the priorities of the first demand and the second demand are intelligently assigned according to the business rule-guided two-way attention demand difference semantic representation between the demand difference semantic parsing features and the business rule semantic parsing features. In this way, the accuracy and rationality of demand priority evaluation can be effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0009] Figure 1 FIG. is a system block diagram of a demand priority intelligent assignment system based on demand semantic parsing according to an embodiment of the present application.

[0010] Figure 2 FIG. is a schematic diagram of data flow of a demand priority intelligent assignment system based on demand semantic parsing according to an embodiment of the present application.

[0011] Figure 3It is a block diagram of a requirement difference two-way evaluation module in a requirement priority intelligent assignment system based on requirement semantic parsing according to an embodiment of the present application. Detailed implementation manners

[0012] Next, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0013] In today's era of rapid digital development, enterprises and organizations face various requirement challenges, including software development, business process optimization, and product iteration. Efficient requirement management has become a core element to ensure project success and business sustainable development. By scientifically determining the priority of requirements, limited resources can be allocated and utilized more reasonably, ensuring that key requirements can be responded to in a timely manner, thereby improving the success rate of projects and the efficiency of resource utilization.

[0014] However, traditional methods for determining requirement priorities, such as relying on experience judgment, simple voting, or only ranking according to customer importance, are difficult to adapt to today's complex and changeable business environment. These methods often lack objectivity and scientific basis, are easily affected by subjective factors, resulting in deviations in priority judgment, and thus have an adverse impact on the project schedule, cost, and quality. In addition, modern business rules are increasingly complex and change rapidly. Traditional technologies face many difficulties in extracting and applying these rules, and it is difficult to effectively combine the determination of requirement priorities with actual business rules, ultimately unable to ensure that the priority assignment can truly meet business requirements.

[0015] To address the above technical problems, the technical concept of the present application is to obtain a first requirement text and a second requirement text, use artificial intelligence-based natural language analysis and coding technologies to perform semantic parsing on the first requirement text and the second requirement text. At the same time, business rules are extracted and semantically parsed. Then, the requirement difference semantic parsing features between the semantic parsing features of the first requirement text and the semantic parsing features of the second requirement text are calculated. Based on this, the priorities of the first requirement and the second requirement are intelligently assigned according to the two-way attention requirement difference semantic representation guided by business rules between the requirement difference semantic parsing features and the business rule semantic parsing features. The present application can perform in-depth semantic parsing on requirement texts, ensuring a more accurate and comprehensive understanding of requirement content, reducing the influence of subjective judgment. Moreover, it closely combines business rules with requirement difference evaluation, fully considering the importance weights of business rules, thereby improving the accuracy and rationality of evaluation.

[0016] Figure 1 It is a system block diagram of a requirement priority intelligent assignment system based on requirement semantic parsing according to an embodiment of the present application. Figure 2The figure is a schematic diagram of data flow of a requirement priority intelligent assignment system based on requirement semantic parsing according to an embodiment of the present application. As Figure 1 and Figure 2 shown, in a requirement priority intelligent assignment system 100 based on requirement semantic parsing, it includes: a requirement text acquisition module 110, configured to acquire a first requirement text and a second requirement text; a requirement text semantic parsing module 120, configured to perform semantic parsing on the first requirement text and the second requirement text to obtain a first requirement text semantic parsing encoded vector and a second requirement text semantic parsing encoded vector; a business rule semantic parsing module 130, configured to extract a business rule and perform semantic parsing on the business rule to obtain a business rule semantic parsing encoded vector; a requirement difference calculation module 140, configured to calculate a requirement difference semantic parsing encoded vector between the first requirement text semantic parsing encoded vector and the second requirement text semantic parsing encoded vector; a requirement difference two-way evaluation module 150, configured to perform a business rule-based requirement difference two-way attention evaluation on the requirement difference semantic parsing encoded vector and the business rule semantic parsing encoded vector to obtain a requirement difference evaluation semantic encoded vector under the business rule; and an assignment result generation module 160, configured to obtain an assignment result based on the requirement difference evaluation semantic encoded vector under the business rule.

[0017] In an embodiment of the present application, the requirement text acquisition module 110 is configured to acquire a first requirement text and a second requirement text. It should be understood that the first requirement text and the second requirement text may include details of specific business requirements, value information that can be brought to the business after completing this requirement, time sensitivity, dependency relationships with other business requirements, and other contents. By performing semantic analysis on the contents of the first requirement text and the second requirement text, the model can understand the business value, influence scope, urgency, etc. of different requirements, and based on this information, determine the requirement priority, and make a more scientific and objective priority decision. In particular, the first requirement text and the second requirement text can be acquired from project requirement documents and customer requirement records.

[0018] In the embodiment of the present application, the requirement text semantic parsing module 120 is used to perform semantic parsing on the first requirement text and the second requirement text to obtain a first requirement text semantic parsing encoded vector and a second requirement text semantic parsing encoded vector. Specifically, in the embodiment of the present application, the requirement text semantic parsing module is used to: use a semantic encoder including a Bert model to perform semantic parsing on the first requirement text and the second requirement text to obtain the first requirement text semantic parsing encoded vector and the second requirement text semantic parsing encoded vector. It should be understood that the first requirement text and the second requirement text contain rich and diverse semantic information, and there is a close semantic association between these information in the context. Based on this, in the technical solution of the present application, by using a semantic encoder including a Bert model to perform semantic parsing on the first requirement text and the second requirement text to capture the complex semantic information and context dependence relationships in the text, a first requirement text semantic parsing encoded vector and a second requirement text semantic parsing encoded vector are obtained. Those of ordinary skill in the art should know that the Bert model is a pre-trained language model, which is based on the Transformer architecture and introduces a bidirectional encoding mechanism. Different from traditional unidirectional language models, the Bert model can consider the left and right context information of a word at the same time, so as to more accurately understand the true meaning of each word in the sentence. That is to say, the requirement text is semantically complex and has context associations. The Bert model uses the multi-head attention mechanism to consider the information of other words when encoding each word, so as to accurately understand the sentence semantics. In this way, the obtained vector can comprehensively reflect the true meaning and internal logic of the requirement text, laying a foundation for further intelligent analysis.

[0019] The following is a detailed elaboration of a specific implementation process of "using a semantic encoder including a Bert model to perform semantic parsing on the first requirement text and the second requirement text to obtain the first requirement text semantic parsing encoded vector and the second requirement text semantic parsing encoded vector": First, it is necessary to select from among many pre-trained versions of the Bert model. Options such as Bert-base-uncased and Bert-large-uncased are all available for consideration. These versions are carefully trained on a vast amount of text data, accumulating rich semantic information and possessing powerful language understanding capabilities. However, which specific version to choose needs to be determined according to the actual requirements and the status of computing resources. For example, if dealing with relatively conventional requirement texts and having limited computing resources, Bert-base-uncased may be a good choice; while for more complex situations with higher requirements for semantic understanding accuracy, Bert-large-uncased may be more suitable.

[0020] Next is the text preprocessing stage. The first step in this stage is tokenization, which is an important operation to convert the text into a form that can be processed by the model. Generally, a tokenizer matching the selected Bert model is used to process the first requirement text and the second requirement text separately. For the first requirement text, assume it is a relatively complex description, such as "It is necessary to develop a new online shopping platform that should have a convenient search function, allowing users to quickly find the products they want through various filtering conditions, support multiple payment methods, and have an intuitive user interface for easy operation by users". The tokenizer will split it into basic elements one by one. It may split this sentence into word or sub-word units like ["It", "is", "necessary", "to", "develop", "a", "new", "online", "shopping", "platform", "that", "should", "have", "a", "convenient", "search", "function", "allowing", "users", "to", "quickly", "find", "the", "products", "they", "want", "through", "various", "filtering", "conditions", "support", "multiple", "payment", "methods", "and", "have", "an", "intuitive", "user", "interface", "for", "easy", "operation", "by", "users"], or even split it into sub-words in a more fine-grained way. This is because the input of the Bert model is based on its pre-trained vocabulary, and this splitting can ensure that the model can accurately process each element in the text. For the second requirement text, no matter what its specific content is, it will also go through the same tokenization operation.

[0021] After tokenization, special tokens need to be added to these texts. This is a requirement of the input format of the Bert model. The [CLS] token will be added at the beginning of each requirement text, which indicates the start of the sentence and enables the Bert model to identify the starting position of the text. The [SEP] token will be added at the end to indicate the end of the sentence or to separate different sentences. For the first requirement text, it will become the form of [CLS] It is necessary to develop a new online shopping platform that should have a convenient search function, allowing users to quickly find the products they want through various filtering conditions, support multiple payment methods, and have an intuitive user interface for easy operation by users[SEP]. The same operation will be performed on the second requirement text. In addition, if the requirement text involves multiple sentences, the [SEP] token will be used to separate different sentences, so that the Bert model can clearly distinguish different sentence structures.

[0022] In addition to special tokens, position encoding and segment encoding are also indispensable parts of the preprocessing. Position encoding plays a crucial role in enabling the model to understand the position information of words in a sentence. It adds an encoding representing the position of each word, allowing the model to know the order of words. For example, in the first requirement text above, "need" might be assigned a position encoding, and "develop" would be assigned another, and so on. Words in different positions will have different position encodings, enabling the model to distinguish the order of words, as word order is crucial for semantic understanding in natural language. Segment encoding is used to distinguish different sentences. When the first requirement text and the second requirement text are used as inputs, they will be marked as belonging to different segments, which helps the Bert model distinguish different input parts.

[0023] After completing these preprocessing tasks, the processed first requirement text and second requirement text can be input into the Bert model. The processing process of the Bert model is a multi-layer architecture processing flow full of wisdom. First, in the word embedding layer, it maps the input word tokens to vector representations. This means that each word token will find its corresponding initial vector representation according to the pre-trained vocabulary of Bert, and this vector contains the original semantic information of the word. Then, these initial vectors are added to the vectors of position encoding and segment encoding to form a comprehensive input vector. Such a comprehensive vector not only contains the semantics of the word but also incorporates the position information of the word in the sentence and the segment information of the sentence, providing a richer information basis for subsequent processing.

[0024] Subsequently, the input vector enters a multi-layer Transformer architecture. This is the core part of the Bert model, and the multi-head self-attention mechanism within it is a key feature. During this process, each word will, through the multi-head self-attention mechanism, focus on the information of other words in the sentence, and at the same time, be focused on by other words. Taking the first requirement text as an example, for the word "search function", during the processing, it will establish connections with other words such as "convenient", "filtering conditions", "users", etc., because there are semantic associations among them. The vector representation of the "search function" will be dynamically adjusted according to its context in the sentence, rather than being viewed in isolation. For example, in the part "Users can quickly find the products they want through various filtering conditions", the semantics of the "search function" will be affected by words such as "filtering conditions" and "finding products". Through the multi-head self-attention mechanism, the "search function" will comprehensively consider its relationships with these words and adjust its vector representation to more accurately reflect its semantics in this context. Moreover, this information exchange is two-way, and "filtering conditions" will also be affected by words such as "search function". This dynamic adjustment will continue in multiple Transformer layers, and the vector representation of each word will be continuously updated and refined. Through the information transfer between layers, semantic information at different levels will be gradually integrated into the final vector representation. For longer requirement texts, this feature of the Bert model is particularly important because it can well capture long-distance semantic dependencies. For example, in a long sentence, the word at the beginning can establish a semantic connection with the word at the end, overcoming the difficulties of traditional language models in dealing with long-distance dependencies.

[0025] After being processed by multiple Transformer layers, output vectors will finally be obtained. For the first requirement text, a semantic parsing encoding vector for the first requirement text will be obtained, and for the second requirement text, a semantic parsing encoding vector for the second requirement text will be obtained. Usually, attention needs to be paid to the output vector corresponding to the [CLS] token at the beginning of each sentence because it is considered to be the semantic representation of the entire sentence and can comprehensively reflect the semantic information of the whole sentence. Taking the first requirement text as an example, the output vector corresponding to the [CLS] token will integrate all the words in the whole sentence and their relationships, containing comprehensive semantic information about various functions required for developing an online shopping platform and user experience requirements. At the same time, according to needs, the average or maximum value of the output vectors of all words in the whole sentence can also be taken as the semantic representation of the sentence, depending on the specific application scenario and different requirements for semantic information. These obtained semantic parsing encoding vectors contain rich semantic information of the input requirement text, not just the simple meanings of words, but also covering the relationships between words and the overall semantics of the sentence.

[0026] In an embodiment of the present application, the business rule semantic parsing module 130 is used to extract business rules and perform semantic parsing on the business rules to obtain business rule semantic parsing encoding vectors. Specifically, in an embodiment of the present application, the business rule semantic parsing module is used to: use the semantic encoder containing the Bert model to perform semantic parsing on the business rules to obtain the business rule semantic parsing encoding vector. It should be understood that business rules contain information related to business strategies and goals, business process specification information, business resource allocation information, etc. In detail, business strategies and goals provide a macro direction for demand priority setting. Demands that meet long-term strategies or can help achieve short-term goals are usually given a higher priority because they directly serve the core business demands of the enterprise and help the enterprise gain competitive advantages or achieve phased results in the market. Business process specifications determine the closeness between demand and business operation. The priority of requirements related to key process steps and time constraints will be adjusted due to the impact on the smoothness and timeliness of business processes. The allocation of business resources limits the possibility and feasibility of demand realization. Demands with sufficient manpower and financial support will have more advantages in priority determination due to better implementation conditions. Considering that the business rules of an enterprise are formulated based on business goals, extracting business rules can make the determination of demand priorities closely revolve around business goals. For example, if the business goal of an enterprise is to improve user satisfaction, extracting related business rules, such as "improving customer service response speed, 90% of inquiries need to be replied within 10 minutes", can make it clear that demands such as "optimizing the intelligent response function of the customer service system" should have a higher priority, ensuring the consistency of demands with business goals, and avoiding waste of resources on demands that are not related to business goals. In addition, business rules are not static and will change with factors such as market environment, industry competition, and corporate strategy adjustments. Extracting business rules can perceive these changes in a timely manner and adjust demand priorities accordingly. Next, considering that business rules often have complex semantic connotations, it is difficult to deeply explore their internal logic in a simple text form, and there are semantic relationships between the various parts of the business rules. Therefore, in order to capture the deep semantic information in these rules and ensure a more accurate and comprehensive understanding of the rules, this application obtains business rule semantic parsing encoding vectors by semantically parsing the business rules. In particular, in a specific embodiment of the present application, the semantic encoder including the Bert model is used to perform semantic analysis on the business rules to utilize the powerful semantic understanding ability of the Bert model to deeply analyze complex business rules, sort out the logical relationship between different conditions of the business rules, accurately grasp the connotation of the rules, and obtain the semantic analysis encoding vector of the business rules, so as to comprehensively reflect the true meaning of the rules and their internal logic, and lay a solid foundation for further intelligent analysis and decision-making.

[0027] In the embodiment of the present application, the requirement difference calculation module 140 is configured to calculate a requirement difference semantic parsing encoding vector between the first requirement text semantic parsing encoding vector and the second requirement text semantic parsing encoding vector. Specifically, in the embodiment of the present application, the requirement difference calculation module is configured to: calculate the position-by-position difference between the first requirement text semantic parsing encoding vector and the second requirement text semantic parsing encoding vector to obtain the requirement difference semantic parsing encoding vector. It should be understood that the first requirement text semantic parsing encoding vector and the second requirement text semantic parsing encoding vector respectively represent the semantic quantization representations of two different requirement texts, and there are different difference information. Therefore, in order to capture and refine the subtle differences between the two at the semantic level, the present application calculates the position-by-position difference between the first requirement text semantic parsing encoding vector and the second requirement text semantic parsing encoding vector to clearly show the differences between the two requirement texts in each semantic detail, and obtain the requirement difference semantic parsing encoding vector, providing data support for subsequent priority quantization. In the process of requirement priority determination, this quantified difference information is an important reference basis. For example, if the differences between two requirement texts in key semantic dimensions are large, it indicates that their differences in functions, goals, etc. are obvious. According to business rules, different priority settings may be required to help the system more objectively and accurately judge the priority of requirements.

[0028] In the embodiment of the present application, the requirement difference two-way evaluation module 150 is configured to perform a business rule-based requirement difference two-way attention evaluation on the requirement difference semantic parsing encoding vector and the business rule semantic parsing encoding vector to obtain a requirement difference evaluation semantic encoding vector under the business rule. Specifically, Figure 3 FIG. is a block diagram of a requirement difference two-way evaluation module in a requirement priority intelligent assignment system based on requirement semantic parsing according to an embodiment of the present application. As Figure 3As shown in the figure, the bidirectional evaluation module 150 for demand differences includes: a projective transformation encoding unit 151, configured to perform a homography projective transformation on the semantic parsing encoding vector of the demand differences and the semantic parsing encoding vector of the business rules to obtain a homography projective encoding vector for semantic parsing of demand differences and a homography projective encoding vector for semantic parsing of business rules; a forward and backward semantic score field calculation unit 152, configured to obtain a forward demand difference - business rule semantic parsing attention score field and a backward demand difference - business rule semantic parsing attention score field based on the homography projective encoding vector for semantic parsing of demand differences and the homography projective encoding vector for semantic parsing of business rules; a bidirectional attention balance field construction unit 153, configured to construct a forward and backward demand difference - business rule semantic parsing bidirectional attention balance field between the forward demand difference - business rule semantic parsing attention score field and the backward demand difference - business rule semantic parsing attention score field; a demand difference - business rule semantic modulation response unit 154, configured to perform modulation interaction on the homography projective encoding vector for semantic parsing of demand differences and the homography projective encoding vector for semantic parsing of business rules based on the forward and backward demand difference - business rule semantic parsing bidirectional attention balance field to obtain a semantic encoding vector for demand difference evaluation under the business rules.

[0029] It should be understood that business rules are the guidelines for business operations, and it is of practical significance to conduct demand difference evaluation within the framework of business rules. For example, in the e - commerce business, the business rules may stipulate that "demands directly related to improving the user purchase conversion rate are given priority". Therefore, in order to perform more accurate priority assignment based on business rules, in the technical solution of this application, a bidirectional attention evaluation mechanism for demand differences based on business rules is introduced to process the semantic parsing encoding vector of the demand differences and the semantic parsing encoding vector of the business rules to obtain a semantic encoding vector for demand difference evaluation under the business rules. In particular, this mechanism can effectively capture the complex semantic associations between the two. For example, demand differences may be reflected in multiple aspects, and different clauses in business rules may have different considerations for these differences. This can help the system, when evaluating, simultaneously focus on the parts of the demand difference vector related to each clause of the business rules and the parts of the business rule vector sensitive to demand differences, so as to more comprehensively and accurately understand the relationship between them.

[0030] Specifically, first, a homography projective transformation is performed on the semantic parsing encoding vector of the demand differences and the semantic parsing encoding vector of the business rules to obtain a homography projective encoding vector for semantic parsing of demand differences and a homography projective encoding vector for semantic parsing of business rules. The above process can be expressed as: ; where is the semantic parsing encoding vector of the demand differences, is the semantic parsing encoding vector of the business rules, is the demand difference mapping homography matrix, is the business rule mapping homography matrix, is the demand difference semantic parsing homography projection coding vector, is the business rule semantic parsing homography projection coding vector.

[0031] It should be understood that the feature space formed by the original demand difference semantic parsing coding vector and the business rule semantic parsing coding vector may only reflect the surface or conventional feature relationships. In order to be able to uncover the potential relationships hidden behind the data and better reveal the deeper associations between requirements and rules, in this application, a homography projection transformation needs to be performed on the demand difference semantic parsing coding vector and the business rule semantic parsing coding vector to map these two coding vectors into a new feature space. In the new feature space, the demand difference semantic parsing homography projection coding vector and the business rule semantic parsing homography projection coding vector obtained after the transformation can present information from different perspectives and scales, which helps the model to more accurately implement demand priority judgment.

[0032] Specifically, in the embodiment of this application, the forward and backward semantic score field calculation unit is used to: calculate the forward demand difference - business rule semantic parsing attention score field of the demand difference semantic parsing homography projection coding vector relative to the business rule semantic parsing homography projection coding vector. This process can be expressed by the formula: ; where, is the demand difference semantic parsing homography projection coding vector, represents matrix multiplication, is the transposed vector of, is the length of, is the forward demand difference - business rule semantic parsing attention score field.

[0033] Calculate the backward demand difference - business rule semantic parsing attention score field of the business rule semantic parsing homography projection coding vector relative to the demand difference semantic parsing homography projection coding vector. This process can be expressed by the formula: ; where, is the business rule semantic parsing homography projection coding vector, is the transposed vector of, is matrix multiplication, is the length of, is the backward demand difference - business rule semantic parsing attention score field.

[0034] It should be understood that during the process of determining the requirement priority, it is necessary to clearly understand the close connection between requirement differences and business rules. By calculating the positive requirement difference - business rule semantic parsing attention score field, the importance distribution of the homography projection coding vector of the requirement difference semantic parsing relative to the homography projection coding vector of the business rule semantic parsing can be quantified. For example, requirement differences may be reflected in aspects such as optimizing product display and simplifying the payment process. Calculating the positive requirement difference - business rule semantic parsing attention score field can clarify the importance of these requirement differences in the context of the business rule of improving the purchase conversion rate, thereby providing a key basis for judging the requirement priority. Moreover, in actual business, requirement differences often contain a lot of information, and not all parts are closely related to business rules. The calculation of the positive requirement difference - business rule semantic parsing attention score field is similar to the focusing mechanism of the human visual system, enabling the model to focus on the part of the requirement difference that best reflects the relevance to business rules. For example, in a software development project, requirement differences may involve various ways of function implementation, different user interface designs, etc. However, the business rule emphasizes the stability and compatibility of the system. By calculating the attention score field, the model can focus on the parts of the requirement difference related to stability and compatibility and ignore the information that has less impact on the business rule, thus more accurately evaluating the degree to which the requirements meet the business rules.

[0035] Correspondingly, considering that relying solely on the positive requirement difference - business rule semantic parsing attention score field may not cover all the associated information between requirement differences and business rules. By calculating the reverse requirement difference - business rule semantic parsing attention score field of the homography projection coding vector of the business rule semantic parsing relative to the homography projection coding vector of the requirement difference semantic parsing, the relationship between the two can be examined from the opposite direction, supplementing the details that may be overlooked in the positive analysis. Moreover, in actual business scenarios, the relationship between requirement differences and business rules is often not symmetric. The positive attention score field reflects the importance of requirement differences in the context of business rules, while the reverse score field can reflect the unique impact of business rules in the context of requirement differences. By calculating the requirement difference - business rule semantic parsing attention score field, this asymmetric relationship can be captured, which helps to improve the accuracy of subsequent requirement priority determination.

[0036] Then, construct the positive and negative requirement difference - business rule semantic parsing two-way attention balance field between the positive requirement difference - business rule semantic parsing attention score field and the reverse requirement difference - business rule semantic parsing attention score field. The above process can be expressed as: ; where is the positive requirement difference - business rule semantic parsing attention score field, is the reverse requirement difference - business rule semantic parsing attention score field, is a feature splicing operation, is a convolution encoding with a 3×3 convolution kernel, is a positive and negative demand difference - business rule semantic parsing two - way attention balance field.

[0037] It should be understood that the separate positive demand difference - business rule semantic parsing attention score field and the negative demand difference - business rule semantic parsing attention score field respectively provide information about the relationship between demand differences and business rules from different directions. However, in order to more comprehensively and accurately evaluate the interactive influence between the two, it is necessary to integrate the information from these two aspects. Based on this, in this application, it is necessary to construct a positive and negative demand difference - business rule semantic parsing two - way attention balance field between the positive demand difference - business rule semantic parsing attention score field and the negative demand difference - business rule semantic parsing attention score field. The constructed positive and negative demand difference - business rule semantic parsing two - way attention balance field ensures that in the final interactive response encoding, the contributions of the positive and negative score fields are reasonably considered, and the importance of the other party will not be ignored due to the excessive influence of a certain score field.

[0038] Specifically, in the embodiment of this application, the demand difference - business rule semantic modulation response unit is used to: map the demand difference semantic parsing homographic projection encoding vector and the business rule semantic parsing homographic projection encoding vector to the positive and negative demand difference - business rule semantic parsing two - way attention balance field respectively to obtain a demand difference semantic parsing homographic projection attention modulation encoding vector and a business rule semantic parsing homographic projection attention modulation encoding vector. This process can be expressed by the formula: ; where, is the demand difference semantic parsing homographic projection encoding vector, is the business rule semantic parsing homographic projection encoding vector, is the positive and negative demand difference - business rule semantic parsing two - way attention balance field, is matrix multiplication, is the demand difference semantic parsing homographic projection attention modulation encoding vector, is the business rule semantic parsing homographic projection attention modulation encoding vector.

[0039] Calculate the element - wise division between the demand difference semantic parsing homographic projection attention modulation encoding vector and the business rule semantic parsing homographic projection attention modulation encoding vector to obtain the demand difference evaluation semantic encoding vector under the business rule. This process can be expressed by the formula: ; where, is the demand difference semantic parsing homographic projection attention modulation encoding vector, It is the homography projection attention modulation coding vector for business rule semantic parsing, which is the semantic coding vector for demand difference evaluation under the said business rule.

[0040] It should be understood that mapping the demand difference semantic parsing homography projection coding vector and the business rule semantic parsing homography projection coding vector to the positive and negative demand difference - business rule semantic parsing bidirectional attention balance field realizes the deep fusion of the information of the two vectors. Specifically, the original coding vectors respectively carry the information related to demand difference and business rules, while the balance field integrates the information of their mutual influence. Through mapping, the newly generated demand difference semantic parsing homography projection attention modulation coding vector and the business rule semantic parsing homography projection attention modulation coding vector not only retain their own original features, but also incorporate the influence brought by the features of the other party. For example, in e-commerce business, the demand difference coding vector may involve various schemes for optimizing product display, and the business rule coding vector focuses on the rules for increasing sales and user satisfaction. After mapping, the new demand difference modulation coding vector will incorporate the requirements of business rules for sales and user satisfaction, thus more comprehensively reflecting the actual meaning of this demand under the business rule framework.

[0041] Correspondingly, considering that in the process of determining demand priority, it is not enough to only understand the respective features of demand difference and business rules. More importantly, it is necessary to clarify the relationship between them. Although the previous steps modulated the demand difference semantic parsing homography projection coding vector and the business rule semantic parsing homography projection coding vector to incorporate the influence of the other party, it is still necessary to further explore the new relationship generated after this fusion. The dot-division operation, as an effective comparison mechanism, can deeply analyze the two modulated feature vectors and reveal the hidden and deep connections between them. Based on this, in this application, it is necessary to calculate the element-wise division between the demand difference semantic parsing homography projection attention modulation coding vector and the business rule semantic parsing homography projection attention modulation coding vector to obtain the semantic coding vector for demand difference evaluation under the business rule. The semantic coding vector for demand difference evaluation under the business rule obtained by the dot-division operation can reveal the subtle proportional relationship between features and provide detailed and crucial basis for subsequent demand priority determination.

[0042] In the embodiment of the present application, the assignment result generation module 160 is configured to obtain an assignment result based on the semantic encoding vector of the requirement difference evaluation under the business rules. Specifically, in the embodiment of the present application, the assignment result generation module is configured to: input the semantic encoding vector of the requirement difference evaluation under the business rules into the requirement priority assignment module based on a classifier to obtain the assignment result, where the assignment result is used to represent that the priority of the first requirement is higher than the priority of the second requirement, the priority of the first requirement is equal to the priority of the second requirement, or the priority of the first requirement is lower than the priority of the second requirement. That is, classification processing is performed on the semantic encoding vector of the requirement difference evaluation under the business rules obtained by two-way evaluation of the requirement difference semantic parsing encoding vector and the business rule semantic parsing encoding vector, so as to intelligently assign the priorities of the first requirement and the second requirement. In particular, the classifier is a mature technology in the field of machine learning and has powerful pattern recognition and classification capabilities. It can extract key features from complex encoded vector data and perform classification according to the learned patterns. For example, classifiers such as support vector machines and random forests perform well in processing high-dimensional data classification tasks, can effectively process the multi-dimensional semantic information in the encoded vector, and accurately judge the relationship between requirement priorities. In this way, based on the clear priority assignment result, the project team can allocate resources more reasonably, including human, material, and time resources. Prioritizing the investment of resources in high-priority requirements can ensure that key requirements are met in a timely manner and improve the overall value and efficiency of the project. In particular, in a specific embodiment of the present application, inputting the semantic encoding vector of the requirement difference evaluation under the business rules into the requirement priority assignment module based on a classifier to obtain the assignment result, where the assignment result is used to represent that the priority of the first requirement is higher than the priority of the second requirement, the priority of the first requirement is equal to the priority of the second requirement, or the priority of the first requirement is lower than the priority of the second requirement, includes: performing fully connected encoding on the semantic encoding vector of the requirement difference evaluation under the business rules using the fully connected layer of the classifier to obtain a fully connected encoding feature vector of the requirement difference evaluation under the business rules; inputting the fully connected encoding feature vector of the requirement difference evaluation under the business rules into the Softmax classification function of the classifier to obtain the probability values of the semantic encoding vector of the requirement difference evaluation under the business rules belonging to each classification label, where the classification labels include those for representing that the priority of one requirement is higher than the priority of the second requirement, for representing that the priority of the first requirement is equal to the priority of the second requirement, and for representing that the priority of the first requirement is lower than the priority of the second requirement; and determining the classification label corresponding to the largest probability value as the assignment result.

[0043] In one example, considering that the semantic parsing encoding vector representing the demand difference represents the parsing semantic differential representation encoding features between the first demand text and the second demand text, and the business rule semantic parsing encoding vector represents the semantic parsing encoding features of the business rule, when performing an interactive response based on the positive and negative attention fields of the features, the difference in the population attributes of the encoding semantic features of the different feature patterns under the source semantics will cause differences in the fairness of the reinforcement levels of the positive and negative attention fields, thereby affecting the inclusiveness of the feature distribution interactive response of the semantic encoding vector for demand difference evaluation under the business rule, and reducing the accuracy of the assignment result obtained by its input to the demand priority assignment module based on the classifier.

[0044] Therefore, in this example, when inputting the semantic encoding vector for demand difference evaluation under the business rule into the demand priority assignment module based on the classifier to obtain the assignment result, the semantic encoding vector for demand difference evaluation under the business rule is optimized. The specific optimization process is as follows: Probability is assigned to each eigenvalue of the semantic encoding vector for demand difference evaluation under the business rule. For example, is activated by a function to obtain the semantic encoding probability vector for demand difference evaluation under the business rule, denoted as .

[0045] Determine the eigenvalue mean and the eigenvalue variance of the semantic encoding probability vector for demand difference evaluation under the business rule, and perform a dot product with the semantic encoding probability vector for demand difference evaluation under the business rule to obtain the first semantic encoding probability boundary limit vector for demand difference evaluation under the business rule and the second semantic encoding probability boundary limit vector for demand difference evaluation under the business rule, where represents the dot product.

[0046] Perform a dot product of the absolute value vector of the dot subtraction between the first semantic encoding probability boundary limit vector for demand difference evaluation under the business rule and the second semantic encoding probability boundary limit vector for demand difference evaluation under the business rule with the logarithm vector of the semantic encoding probability vector for demand difference evaluation under the business rule with base 2 to obtain the semantic encoding probability information transfer vector for demand difference evaluation under the business rule, where represents the dot subtraction.

[0047] After performing a dot product of the reciprocal of each bit of the first semantic encoding probability boundary limit vector for demand difference evaluation under the business rule with the second semantic encoding probability boundary limit vector for demand difference evaluation under the business rule, perform a dot addition with the square root of the quotient of the eigenvalue mean divided by the eigenvalue standard deviation to obtain the semantic encoding multi-granularity fairness collaborative vector for demand difference evaluation under the business rule , wherein, represents the reciprocal of each bit of the probability boundary limit vector for demand difference evaluation semantic coding under the first service rule, and the reciprocal of each bit is to calculate the reciprocal of the eigenvalue of each position of the vector. represents dot addition.

[0048] Dot-add the demand difference evaluation semantic coding probability information transfer vector under the service rule and the demand difference evaluation semantic coding multi-granularity fair cooperation vector under the service rule to obtain an optimized demand difference evaluation semantic coding vector under the service rule. .

[0049] That is, considering the dynamic interaction pattern differences existing in the demand difference evaluation semantic coding vector under the service rule, in order to improve the dynamic response fault tolerance rate of the demand difference evaluation semantic coding vector under the service rule under the condition of representing the distribution dispersion degree, a two-way fairness index based on mixed probability boundary restriction is used as the interaction constraint target, the population feature information transfer process of the demand difference evaluation semantic coding vector under the service rule is optimized and adjusted, and by introducing a hierarchical fairness calibration factor based on the global statistical response mechanism, a robust distribution consistency representation system under the multi-granularity fair cooperation architecture of the demand difference evaluation semantic coding vector under the service rule is constructed to form a fairness cooperation framework with self-adaptive adjustment ability, effectively balance the multi-dimensional interaction relationship of the demand difference evaluation semantic coding vector in the feature space, and improve the accuracy of the assignment result obtained by its input to the demand priority assignment module based on the classifier.

[0050] In summary, the demand priority intelligent assignment system 100 based on demand semantic parsing according to the embodiments of the present application is clarified. It uses artificial intelligence-based natural language analysis and coding technologies to perform semantic parsing on the first demand text and the second demand text. At the same time, service rules are extracted and semantic parsing is performed on them. Then, the demand difference semantic parsing features between the semantic parsing features of the first demand text and the semantic parsing features of the second demand text are calculated, and based on this, the priorities of the first demand and the second demand are intelligently assigned according to the two-way attention demand difference semantic representation guided by the service rule between the demand difference semantic parsing features and the service rule semantic parsing features. In this way, the accuracy and rationality of demand priority evaluation can be effectively improved.

Claims

1. A requirement priority intelligent assignment system based on requirement semantic parsing, characterized in that Including: A requirement text acquisition module, configured to acquire a first requirement text and a second requirement text; A requirement text semantic parsing module, configured to perform semantic parsing on the first requirement text and the second requirement text to obtain a first requirement text semantic parsing encoded vector and a second requirement text semantic parsing encoded vector; A business rule semantic parsing module, configured to extract business rules and perform semantic parsing on the business rules to obtain a business rule semantic parsing encoded vector; A requirement difference calculation module, configured to calculate the position-wise difference between the first requirement text semantic parsing encoded vector and the second requirement text semantic parsing encoded vector to obtain a requirement difference semantic parsing encoded vector; A requirement difference two-way evaluation module, configured to perform a business rule-based requirement difference two-way attention evaluation on the requirement difference semantic parsing encoded vector and the business rule semantic parsing encoded vector to obtain a requirement difference evaluation semantic encoded vector under the business rule; An assignment result generation module, configured to obtain an assignment result based on the requirement difference evaluation semantic encoded vector under the business rule; Wherein, the assignment result generation module is configured to: input the requirement difference evaluation semantic encoded vector under the business rule into a requirement priority assignment module based on a classifier to obtain the assignment result, and the assignment result is used to indicate that the priority of the first requirement is higher than the priority of the second requirement, the priority of the first requirement is equal to the priority of the second requirement, or the priority of the first requirement is lower than the priority of the second requirement.

2. The requirement priority intelligent assignment system based on requirement semantic parsing according to claim 1, characterized in that, The requirement text semantic parsing module is configured to: use a semantic encoder including a Bert model to perform semantic parsing on the first requirement text and the second requirement text to obtain the first requirement text semantic parsing encoded vector and the second requirement text semantic parsing encoded vector.

3. The requirement priority intelligent assignment system based on requirement semantic parsing according to claim 2, characterized in that, The business rule semantic parsing module is configured to: use the semantic encoder including the Bert model to perform semantic parsing on the business rules to obtain the business rule semantic parsing encoded vector.

4. The requirement priority intelligent assignment system based on requirement semantic parsing according to claim 3, characterized in that The requirement difference two-way evaluation module includes: A projective transformation encoding unit, configured to perform a homography projective transformation on the requirement difference semantic parsing encoded vector and the business rule semantic parsing encoded vector to obtain a requirement difference semantic parsing homography projective encoded vector and a business rule semantic parsing homography projective encoded vector; A forward and backward semantic score field calculation unit, configured to obtain a forward requirement difference - business rule semantic parsing attention score field and a backward requirement difference - business rule semantic parsing attention score field based on the requirement difference semantic parsing homography projective encoded vector and the business rule semantic parsing homography projective encoded vector; A two-way attention balance field construction unit, configured to construct a forward and backward requirement difference - business rule semantic parsing two-way attention balance field between the forward requirement difference - business rule semantic parsing attention score field and the backward requirement difference - business rule semantic parsing attention score field; The demand difference - business rule semantic modulation response unit is used to modulate and interact the demand difference semantic parsing homography projection coding vector and the business rule semantic parsing homography projection coding vector based on the positive and negative demand difference - business rule semantic parsing bidirectional attention balance field to obtain the demand difference evaluation semantic coding vector under the business rule.

5. The requirement priority intelligent assignment system based on requirement semantic parsing according to claim 4, characterized in that, The forward and reverse semantic score field calculation unit is used for: Calculating the forward demand difference - business rule semantic parsing attention score field of the demand difference semantic parsing homography projection coding vector relative to the business rule semantic parsing homography projection coding vector; Calculating the reverse demand difference - business rule semantic parsing attention score field of the business rule semantic parsing homography projection coding vector relative to the demand difference semantic parsing homography projection coding vector.

6. The requirement priority intelligent assignment system based on requirement semantic parsing according to claim 5, characterized in that, The demand difference - business rule semantic modulation response unit is used for: Mapping the demand difference semantic parsing homography projection coding vector and the business rule semantic parsing homography projection coding vector to the positive and negative demand difference - business rule semantic parsing bidirectional attention balance field respectively to obtain the demand difference semantic parsing homography projection attention modulation coding vector and the business rule semantic parsing homography projection attention modulation coding vector; Calculating the element - wise division between the demand difference semantic parsing homography projection attention modulation coding vector and the business rule semantic parsing homography projection attention modulation coding vector to obtain the demand difference evaluation semantic coding vector under the business rule.

Images