A method and system for processing a computing resource session response applied to deep learning

Abstract

The application discloses a kind of applied to the computing power resource session response processing method and system of deep learning.The application can accurately extract the key information in session by identifying target session interaction knowledge and constructing information set for computing power resource session data, to avoid the interference of irrelevant information.Initial associated session interaction knowledge vector is obtained by encoding and cross-collision on associated session interaction knowledge information set of computing power resource business keywords, which can deeply mine the knowledge connotation related to business keywords and improve the understanding ability of knowledge in different business scenarios.Target associated session interaction knowledge vector can accurately match the most relevant knowledge resources to improve the accuracy of response according to the correlation eigenvalue.Determined session interaction business theme keyword can quickly and accurately respond to computing power resource session to obtain session response result, effectively improve the session processing efficiency, improve the quality of response, and further optimize the management and utilization of computing power resources.

Description

Technical Field

[0001] This application belongs to the field of data analysis technology, specifically relating to a method and system for processing computing resource session responses applied to deep learning. Background Technology

[0002] In the current field of computing resource management, with the widespread application of computing resources, session processing faces numerous challenges. Traditional session response methods often fail to effectively extract key information from complex session data, leading to inaccurate understanding of computing resource business needs. For example, when allocating computing resources to an enterprise, a user's session may contain multiple requirements, but traditional methods struggle to accurately identify and associate them with specific business keywords. Furthermore, the lack of methods for in-depth mining of knowledge related to business keywords prevents the full utilization of existing knowledge resources during responses, significantly compromising the accuracy and comprehensiveness of the responses. Moreover, existing technologies are inefficient in associating and matching session knowledge with business keywords, resulting in slow session response speeds and impacting the efficiency of computing resource management. Summary of the Invention

[0003] This application provides a method and system for processing computing resource session responses for deep learning, which can solve or partially solve the technical problems involved in the background art mentioned above.

[0004] This application provides a method for processing computing resource session responses in deep learning, applied to a session response processing system. The method includes: acquiring computing resource session data to be processed; identifying target session interaction knowledge in the computing resource session data to be processed to obtain a target session interaction knowledge information set; mining the current session interaction knowledge vector corresponding to the target session interaction knowledge information set; and obtaining the initial associated session interaction knowledge vector corresponding to each computing resource business keyword. The initial associated session interaction knowledge vector is obtained by encoding each associated session interaction knowledge information set corresponding to the computing resource business keyword into a knowledge vector, resulting in each associated session interaction knowledge encoding. The interaction knowledge vector is obtained by cross-collision. Based on the correlation feature values ​​of the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing resource business keyword, the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector is determined from the associated session interaction knowledge vectors corresponding to each computing resource business keyword. The computing resource business keyword corresponding to the target associated session interaction knowledge vector is used as the session interaction business topic keyword of the corresponding target session interaction knowledge information set. Based on the session interaction business topic keyword of the target session interaction knowledge information set, the computing resource session response result corresponding to the computing resource session data to be processed is determined.

[0005] This application provides a session response processing system, including at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the method described above.

[0006] This application provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the above-described method.

[0007] This application identifies target session interaction knowledge and constructs information sets from computing resource session data, enabling precise extraction of key information from the session and avoiding interference from irrelevant information. Encoding and cross-colliding the associated session interaction knowledge information sets of computing resource business keywords yields an initial associated session interaction knowledge vector, allowing for in-depth mining of the knowledge connotations related to business keywords and improving the understanding of knowledge in different business scenarios. Determining the target associated session interaction knowledge vector based on relevance feature values ​​accurately matches the most relevant knowledge resources to improve response accuracy. Deriving session response results based on the determined session interaction business topic keywords enables rapid and accurate responses to computing resource sessions, effectively improving session processing efficiency and response quality, thereby optimizing the management and utilization of computing resources. Attached Figure Description

[0008] Figure 1 A flowchart of a computing resource session response processing method for deep learning provided in this application.

[0009] Figure 2 This is a schematic diagram of the structure of a session response processing system provided in this application. Detailed Implementation

[0010] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, in this application, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects have an "or" relationship.

[0012] Figure 1 A method for processing computing resource session responses for deep learning is shown. The method is applied to a session response processing system and includes the following steps 202-208.

[0013] Step 202: Obtain the computing power resource session data to be processed, identify the target session interaction knowledge in the computing power resource session data to be processed, and obtain the target session interaction knowledge information set.

[0014] Step 204: Mine the current session interaction knowledge vector corresponding to the target session interaction knowledge information set, and obtain the initial associated session interaction knowledge vector corresponding to each computing power resource business keyword.

[0015] In this application, the initial associated session interaction knowledge vector is obtained by encoding each associated session interaction knowledge information set corresponding to the computing power resource business keywords into knowledge vectors, obtaining each associated session interaction knowledge code, and then cross-colliding the various associated session interaction knowledge codes.

[0016] Step 206: Based on the correlation feature values ​​of the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing power resource business keyword, determine the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector from the associated session interaction knowledge vectors corresponding to each computing power resource business keyword.

[0017] Step 208: Use the computing resource business keywords corresponding to the target associated session interaction knowledge vector as the session interaction business topic keywords of the corresponding target session interaction knowledge information set, and determine the computing resource session response result corresponding to the computing resource session data to be processed based on the session interaction business topic keywords of the target session interaction knowledge information set.

[0018] In this application, the technical solutions described in steps 202-208 above are illustrated by the following exemplary description.

[0019] In step 202, the data from the computing resource sessions to be processed comes from a wide range of sources, including user-computing resource service provider interfaces, such as user consultation sessions on cloud computing service platforms or internal communication records between employees regarding the allocation and use of computing resources. This data can be in text format and includes various questions, descriptions, and requests related to computing resources. For example, the data might be something like, "Our department needs to perform large-scale data mining tasks; how much computing power is appropriate?" or "The computing power of the GPU server suddenly drops during deep learning model training; how do we troubleshoot the cause?"

[0020] When acquiring data, it's crucial to ensure its integrity and accuracy. Data from different sources may require format standardization and initial cleaning. For example, unifying text with different encoding formats to UTF-8 and removing garbled characters and irrelevant special characters.

[0021] To identify knowledge from target conversational interactions, knowledge extraction methods from Natural Language Processing (NLP) can be employed. For example, rule-based methods can be used to predefine grammatical rules and keyword patterns related to computing resources. Taking the identification of knowledge about computing resource capacity as an example, the rule could be: "If keywords such as 'need,' 'computing resources,' and 'how much' appear, followed by expressions related to numbers or capacity units (such as GB, TFLOPS, etc.), then this is extracted as knowledge about computing resource capacity requirements."

[0022] Alternatively, classification algorithms from machine learning, such as Support Vector Machines (SVM), can be used to classify and identify knowledge in conversation data. First, a labeled training set needs to be constructed, containing a large number of labeled samples of computing resource conversation data. For example, statements about computing resource performance can be labeled as one category, and statements about computing resource costs as another. Then, an SVM model is trained using this training set. The conversation data to be processed is input into the trained SVM model to obtain classification results, thereby identifying different types of target conversation interaction knowledge.

[0023] After identification, the relevant knowledge is organized into a target session interaction knowledge information set. For example, an information set may contain content such as {"Computing resource requirement type: large-scale data mining task", "Involved equipment: GPU server", "Problem type: computing resource requirement assessment"}.

[0024] Next, in step 204, to mine the current conversational interaction knowledge vector corresponding to the target conversational interaction knowledge information set, a word vector model, such as Word2Vec or GloVe, can be used. Taking Word2Vec as an example, each word in the target conversational interaction knowledge information set is first mapped to a low-dimensional vector space. For example, if the vocabulary size is (V) and the vector dimension is (d) (e.g., (d = 300)), then each word is represented as a (d)-dimensional vector.

[0025] For a target session interaction knowledge information set, the word vectors within it are combined. An exemplary combination method is to calculate the average vector. For instance, if the information set contains three words (w_1), (w_2), and (w_3), with corresponding word vectors (vec{v}_1), (vec{v}_2), and (vec{v}_3) respectively, then the current session interaction knowledge vector (vec{v}_{current}) can be represented as:

[0026] (vec{v}_{current}=frac{vec{v}_1+vec{v}_2+vec{v}_3}{3}).

[0027] For each computing power resource business keyword, the first step is to determine its associated session interaction knowledge information set. For example, for the business keyword "GPU computing power", its associated session interaction knowledge information set may include knowledge content from all historical session data related to GPU computing power, such as "the advantages of GPU computing power in image rendering" and "methods to improve GPU computing power".

[0028] Then, knowledge vector encoding is performed on these sets of related conversational interaction knowledge. For example, a neural network-based encoding method, such as a multilayer perceptron (MLP), can be used. For each set of related conversational interaction knowledge, the word sequence is input into an MLP with multiple hidden layers (e.g., 3 hidden layers, with 128, 64, and 32 neurons per layer, respectively), resulting in a fixed-length (e.g., 128-dimensional) related conversational interaction knowledge encoding.

[0029] Finally, the encodings of each associated session interaction knowledge are cross-collided to obtain the initial associated session interaction knowledge vector. Cross-colliding can be performed using methods such as vector dot product or cosine similarity calculation. For example, for two associated session interaction knowledge encodings (vec{e}_1) and (vec{e}_2), their cross-collision result (taking cosine similarity as an example) is: (cosine(vec{e}_1, vec{e}_2)

[0030] =frac{vec{e}_1cdotvec{e}_2}{vertvec{e}_1vertvertvec{e}_2vert}), and by comprehensively processing multiple such results, the initial associated session interaction knowledge vector is obtained.

[0031] Step 206 involves calculating the relevance feature values ​​between the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing resource business keyword. Various methods can be used here, such as the Pearson correlation coefficient or deep learning-based similarity calculation methods. Taking the Pearson correlation coefficient as an example:

[0032] For the current session interaction knowledge vector (vec{v}_{current}) and the associated session interaction knowledge vector (vec{v}_{associated}) corresponding to a certain computing resource business keyword, the Pearson correlation coefficient (r) is calculated as follows: (r=frac{sum_{i=1}^{n}(x_i-bar{x})(y_i-bar{y})}{sqrt{sum_{i=1}^{n}(x_i-bar{x})^2sum_{i=1}^{n}(y_i-bar{y})^2}}), where (x_i) and (y_i) are the (i)th elements of (vec{v}_{current}) and (vec{v}_{associated}), respectively, (bar{x}) and (bar{y}) are their mean values, and (n) is the dimension of the vector.

[0033] For example, the current session interaction knowledge vector (vec{v}_{current}) is a 100-dimensional vector, and a certain associated session interaction knowledge vector (vec{v}_{associated}) is also a 100-dimensional vector. The Pearson correlation coefficient between them is calculated to be 0.8 using the formula above. This 0.8 is the correlation feature value between them.

[0034] Next, based on the calculated relevance feature values, the target related session interaction knowledge vector corresponding to the current session interaction knowledge vector can be determined from the related session interaction knowledge vectors corresponding to the business keywords of each computing power resource. A threshold can be set; for example, related session interaction knowledge vectors with a relevance feature value greater than 0.7 are determined as target related session interaction knowledge vectors. If multiple related session interaction knowledge vectors meet this condition, the vector with the highest relevance feature value may be selected as the target related session interaction knowledge vector.

[0035] Finally, in step 208, the computing resource business keywords corresponding to the target associated session interaction knowledge vector are used as the session interaction business topic keywords of the corresponding target session interaction knowledge information set. For example, if the computing resource business keyword corresponding to the target associated session interaction knowledge vector is "GPU computing power optimization", then "GPU computing power optimization" is the session interaction business topic keyword of the target session interaction knowledge information set.

[0036] Based on the keywords of the business topic of the target session interaction knowledge information set, determine the computing resource session response result corresponding to the computing resource session data to be processed. This can be achieved using knowledge base query or template-based response generation methods.

[0037] For the knowledge base query method, the system maintains a knowledge base containing a large amount of knowledge related to computing resources. Using the keywords of the conversational interaction business topic as an index, the system searches for corresponding response content in the knowledge base. For example, if the conversational interaction business topic keyword is "GPU computing power optimization," the knowledge base might store response content such as "GPU computing power can be improved by optimizing algorithms, upgrading drivers, and increasing parallel computing."

[0038] For template-based response generation methods, some response templates corresponding to different session interaction business topic keywords are predefined. For example, for the topic keyword "GPU computing power optimization", the template might be "Regarding your question about GPU computing power optimization, here are some suggestions: [List of optimization measures]". Then, the [List of optimization measures] section of the template is filled in according to the specific knowledge content, such as "Regarding your question about GPU computing power optimization, here are some suggestions: optimize algorithm structure, upgrade to the latest GPU driver, and reasonably set the number of parallel computing threads, etc.", thereby obtaining the computing power resource session response result.

[0039] Through the four main steps described above—acquiring the session data of computing resources to be processed, mining relevant knowledge vectors, determining the target-related session interaction knowledge vectors, and finally obtaining the computing resource session response results—an effective computing resource session response processing system has been implemented. This system helps improve the interaction efficiency in computing resource management, enhances the user experience, and can accurately handle various session requirements related to computing resources.

[0040] In summary, this application first identifies target session interaction knowledge and constructs an information set from computing resource session data, enabling precise extraction of key information from the session and avoiding interference from irrelevant information. During the knowledge vector mining process, the associated session interaction knowledge information set of computing resource business keywords is encoded and cross-collided to obtain an initial associated session interaction knowledge vector. This method can deeply mine the knowledge connotations related to business keywords, improving the understanding of knowledge in different business scenarios. Determining the target associated session interaction knowledge vector based on relevance feature values ​​accurately matches the most relevant knowledge resources, thereby improving the accuracy of responses. Finally, based on the determined session interaction business topic keywords, the session response result is obtained, enabling rapid and accurate responses to computing resource sessions, effectively improving session processing efficiency, reducing processing time, and improving response quality, thereby optimizing the management and utilization of computing resources.

[0041] In practical applications, before acquiring the computing power resource session data to be processed, identifying the target session interaction knowledge in the computing power resource session data to be processed, and obtaining the target session interaction knowledge information set, the method further includes: acquiring each associated session interaction knowledge information set corresponding to the set computing power resource business keyword; encoding each associated session interaction knowledge information set into a knowledge vector to obtain each associated session interaction knowledge encoding; and performing cross-collision on each associated session interaction knowledge encoding to obtain the initial associated session interaction knowledge vector corresponding to the set computing power resource business keyword.

[0042] Based on the above, the step of cross-colliding the various associated session interaction knowledge codes to obtain the initial associated session interaction knowledge vector corresponding to the set computing resource business keyword includes: combining the various associated session interaction knowledge codes to obtain a session interaction knowledge combination code; and performing full connection processing based on the session interaction knowledge combination code to obtain the initial associated session interaction knowledge vector corresponding to the set computing resource business keyword.

[0043] Understandably, the first step in the entire computing resource management and session response system is to clearly define computing resource business keywords. These keywords are terms or phrases closely related to computing resource business, such as "GPU computing power optimization," "CPU resource allocation strategy," and "data center computing power load balancing." These keywords form the basis for the entire system to classify and process computing resource knowledge.

[0044] For each defined computing resource business keyword, it is necessary to acquire the corresponding set of related conversation interaction knowledge information. This process involves mining and organizing a large amount of historical conversation data. For example, for the keyword "GPU computing power optimization," the set of related conversation interaction knowledge information may include all historical conversation content related to GPU computing power optimization. This content may include questions raised by users about how to improve the efficiency of GPU computing power in specific tasks (such as deep learning model training), and the corresponding answers provided by technical personnel. These information sets can be obtained from the enterprise's knowledge base, customer service records, or past technical communication documents. During the acquisition process, it is necessary to ensure the completeness and accuracy of the information, which may require data cleaning operations to remove erroneous information, incomplete records, or irrelevant information.

[0045] To transform the knowledge information set from related conversational interactions into a computable and analyzable form, a knowledge vector encoding method is employed. One exemplary approach is neural network-based encoding, such as using a multilayer perceptron (MLP).

[0046] For example, a set of knowledge information related to conversational interactions can be represented as a sequence of words (S = {w_1, w_2, cdots, w_n}), where (w_i) represents the (i)th word in the information set. First, each word needs to be mapped to a low-dimensional vector space. Pre-trained word vector models such as Word2Vec or GloVe can be used. For example, if the vocabulary size is (V) and the word vector dimension is (d) (e.g., (d = 300)), then each word (w_i) is represented as a (d)-dimensional vector (vec{v}_{w_i}).

[0047] Then, these word vectors are sequentially input into an MLP with multiple hidden layers, following the order of the words in the information set. For example, the MLP has three hidden layers, with the number of neurons in each layer being (h_1 = 128), (h_2 = 64), and (h_3 = 32), respectively. The number of neurons in the input layer is equal to the word vector dimension (d), and the number of neurons in the output layer is also a fixed value (e) (e.g., (e = 128)). For the input word vector sequence, a fixed-length (e) dimensional vector is calculated through forward propagation to encode the associated conversational interaction knowledge.

[0048] For the computation from the input layer to the first hidden layer, for example, if the input vector is (vec{x}) (where (vec{x}) is a vector sequence composed of word vectors), the weight matrix of the first hidden layer is (W_1), and the bias vector is (vec{b}_1), then the output (vec{h}_1) of the first hidden layer is: (vec{h}_1=f(W_1vec{x}+vec{b}_1)), where (f) is the activation function, such as the ReLU function (f(x)=max(0,x)).

[0049] Similarly, for the second hidden layer, the input is (vec{h}_1), the weight matrix is ​​(W_2), the bias vector is (vec{b}_2), and the output (vec{h}_2) is: (vec{h}_2=f(W_2vec{h}_1+vec{b}_2)).

[0050] For the third hidden layer, the input is (vec{h}_2), the weight matrix is ​​(W_3), the bias vector is (vec{b}_3), and the output (vec{h}_3) is: (vec{h}_3=f(W_3vec{h}_2+vec{b}_3)).

[0051] Finally, the output of the output layer (vec{y}) (i.e., the encoding of the associated conversational interaction knowledge) is: (vec{y}=f(W_4vec{h}_3+vec{b}_4)), where (W_4) is the weight matrix of the output layer and (vec{b}_4) is the bias vector.

[0052] Furthermore, the associated session interaction knowledge codes are cross-collided to obtain the initial associated session interaction knowledge vector corresponding to the set computing resource business keyword. For example, for a certain set computing resource business keyword, (m) associated session interaction knowledge codes (vec{e}_1, vec{e}_2, cdots, vec{e}_m) are obtained. An exemplary combination method is to concatenate these vectors to obtain the session interaction knowledge combination code (vec{C}). For example, if each associated session interaction knowledge code is a 128-dimensional vector, then the concatenated session interaction knowledge combination code (vec{C}) has a dimension of (128m).

[0053] A fully connected neural network is used to encode and process the combined knowledge of conversational interactions. For example, the input to the push discriminant branch is (vec{C}), the weight matrix is ​​(W_f), the bias vector is (vec{b}_f), and the output is the initial associated conversational interaction knowledge vector (vec{v}). The push discriminant branch is calculated as: (vec{v}=f(W_fvec{C}+vec{b}_f)), where (f) can still be a ReLU activation function. Here, the dimension of the weight matrix (W_f) depends on the dimension requirements of the input vector (vec{C}) and the output vector (vec{v}). For example, if (vec{C}) is (128m) dimensions and the output vector (vec{v}) is required to be 64 dimensions, then the dimension of (W_f) is (64*128m).

[0054] This design, firstly, provides a comprehensive knowledge foundation for subsequent session response processing by pre-acquiring the knowledge information sets of various related session interactions corresponding to the set computing resource business keywords. During the knowledge vector encoding process, neural network methods such as multilayer perceptrons effectively transform the complex set of related session interaction knowledge information into knowledge codes with fixed dimensions, facilitating subsequent calculations and analysis. Combining and fully connecting the various related session interaction knowledge codes yields the initial related session interaction knowledge vector. This method fully integrates information from different related session interaction knowledge codes, enabling the initial related session interaction knowledge vector to more comprehensively reflect the knowledge characteristics related to the set computing resource business keywords. This helps to more accurately match target session interaction knowledge in subsequent session interactions, improving the accuracy and efficiency of computing resource session responses and optimizing the session processing flow in the entire computing resource management process.

[0055] Based on acquiring the knowledge information sets of each associated session interaction corresponding to the set computing resource business keyword; encoding the knowledge information sets of each associated session interaction to obtain the knowledge codes of each associated session interaction; and performing cross-collision on the knowledge codes of each associated session interaction to obtain the initial associated session interaction knowledge vector corresponding to the set computing resource business keyword, the method further includes: loading the knowledge information sets of each associated session interaction into an initial AI knowledge embedding network; encoding the knowledge information sets of each associated session interaction into knowledge vectors through the knowledge vector encoding module in the initial AI knowledge embedding network to obtain the knowledge codes of each associated session interaction; and performing cross-collision on the knowledge codes of each associated session interaction through the knowledge cross-collision module in the initial AI knowledge embedding network to obtain the initial associated session interaction knowledge vector corresponding to the set computing resource business keyword.

[0056] Based on the aforementioned technical solution of the initial AI knowledge embedding network, the debugging method of the initial AI knowledge embedding network includes: acquiring each computing resource session data sample and its corresponding prior session response training annotation, and loading each computing resource session data sample into the original deep learning network; encoding each computing resource session data sample into a knowledge vector using the initial AI knowledge embedding network in the original deep learning network to obtain each computing resource session knowledge vector sample; performing cross-collision on the each computing resource session knowledge vector sample using the initial knowledge cross-collision module in the original deep learning network to obtain a cross-collision sample vector; and outputting the initial session response through the original deep learning network. The module outputs a session response from the cross-collision sample vector to obtain a session response prediction result; it optimizes the original deep learning network based on the prior session response training annotation and the session response prediction result to obtain an intermediate deep learning network; it uses the intermediate deep learning network as the original deep learning network and jumps to the steps of obtaining session data samples of each computing resource and the corresponding prior session response training annotation, and loading the session data samples of each computing resource into the original deep learning network, repeating this process until debugging is completed to obtain the target deep learning network; and it obtains the initial AI knowledge embedding network based on the target knowledge vector encoding module and the target knowledge cross-collision module in the target deep learning network.

[0057] Based on the previously acquired knowledge sets of related conversations corresponding to the set computing resource business keywords, these knowledge sets are loaded into the initial AI knowledge embedding network. This initial AI knowledge embedding network is a deep learning network structure specifically built to process knowledge related to computing resource conversations.

[0058] The knowledge vector encoding module in the initial AI knowledge embedding network is responsible for encoding knowledge vectors for each set of related conversational interaction knowledge information. Various deep learning algorithms can be used here, such as encoding methods based on convolutional neural networks (CNNs).

[0059] For example, the knowledge information set related to conversational interactions can be represented as a two-dimensional matrix (M), where the rows of the matrix represent different word sequences (e.g., each word sequence corresponds to a sentence), and the columns represent the indices of the words in the vocabulary. Let the size of the vocabulary be (V), and the maximum length of each word sequence be (L). For CNN encoding, multiple convolutional kernels are used for convolution operations. Let the kernel size be (k*d) (e.g., (k=3), (d) is the word vector dimension, e.g., (d=300)), and the stride be (s) (e.g., (s=1)).

[0060] For a convolution kernel (w), the formula for convolving a matrix (M) is: (C_{i,j}=(w*M)_{i,j}=sum_{m=0}^{k-1}sum_{n=0}^{d-1}w_{m,n}M_{i+m,j+n}), where (C) is the convolution result matrix. After convolution operations with multiple kernels, pooling operations (such as max pooling) are used to reduce the data dimensionality, resulting in a fixed-length knowledge vector encoding. For example, max pooling is used to take the maximum value within a local region, ultimately obtaining the knowledge encoding of each associated session interaction.

[0061] Next, the knowledge cross-collision module in the initial AI knowledge embedding network performs cross-collision on the obtained related conversational interaction knowledge codes. For example, if (m) related conversational interaction knowledge codes (vec{e}_1, vec{e}_2, cdots, vec{e}_m) are obtained, a method based on matrix multiplication and nonlinear transformation can be used.

[0062] First, construct a matrix (E = [vec{e}_1, vec{e}_2, cdots, vec{e}_m]), then perform matrix multiplication (E*E^T) to obtain an (m*m) matrix (P), where (P_{ij} = vec{e}_icdotvec{e}_j) (where (cdot) represents the vector dot product). Next, perform a nonlinear transformation on each element of matrix (P), for example, using the sigmoid function (f(x) = frac{1}{1+e^{-x}}), to obtain the transformed matrix (P'). Finally, combine the elements of matrix (P') (e.g., by averaging or weighted summation) to obtain the initial associated session interaction knowledge vector corresponding to the set computing power resource business keywords.

[0063] Based on the above, the debugging methods for the initial AI knowledge embedding network include the following.

[0064] Obtaining Data Samples and Training Annotations: First, obtain data samples for each computing resource session and corresponding prior session response training annotations. These data samples are representative samples extracted from actual historical computing resource session data, including session content such as different types of computing resource request inquiries and troubleshooting. The prior session response training annotations are the correct responses to these data samples, provided by experts or verified standard answers.

[0065] Knowledge vector encoding yields session knowledge vector samples: Session data samples from various computing resources are loaded into the initial AI knowledge embedding network within the original deep learning network. These data samples are then encoded using the knowledge vector encoding module within the initial AI knowledge embedding network. Similar to the encoding method mentioned earlier, a specific algorithm is used to transform each session data sample into a knowledge vector form. For example, for a session data sample sentence containing (n) words, after encoding, a fixed-length (e.g., (d' = 128)) computing resource session knowledge vector sample (vec{v}_{s}) is obtained.

[0066] Cross-collision yields a cross-collision sample vector: The initial knowledge cross-collision module in the original deep learning network is used to cross-collide the various computing resource session knowledge vector examples. For example, (p) computing resource session knowledge vector examples (vec{v}_{s1}, vec{v}_{s2}, cdots, vec{v}_{sp}) are obtained, and the cross-collision sample vector (vec{v}_{c}) is obtained according to the cross-collision method mentioned above (such as matrix multiplication and nonlinear transformation).

[0067] The conversation response output yields the prediction result: The initial conversation response output module in the original deep learning network outputs the conversation response from the cross-collision sample vector. This module can be a classifier or generator based on a multilayer perceptron (MLP). For example, if it is a classifier, for (q) possible response types (such as response categories for different types of computing resource problems), the input of the MLP is the cross-collision sample vector (vec{v}_c}), and the output is a (q)-dimensional vector (vec{y}), representing the probability of each response type. The softmax function (y_i=frac{e^{z_i}}{sum_{j=1}^{q}e^{z_j}}) (where (z_i) is the unnormalized value of the output of the last layer of the MLP) can be used to convert the output into a probabilistic form, and the category with the highest probability is the conversation response prediction result.

[0068] Optimizing the Deep Learning Network Based on Annotations: The original deep learning network is optimized based on prior session response training annotations and session response prediction results. A loss function can be used to measure the difference between the prediction result and the correct result. For example, for classification problems, an exemplary cross-entropy loss function is (L=-sum_{i=1}^{q}y_i^{true}log(y_i^{pred})), where (y_i^{true}) is the true probability of the correct response type in the prior session response training annotations (if it is a one-hot encoding, only the correct type is 1, and others are 0), and (y_i^{pred}) is the probability in the session response prediction result. Through the backpropagation algorithm, the parameters in the original deep learning network are adjusted according to the loss function to obtain an intermediate deep learning network.

[0069] The target deep learning network is obtained through iterative debugging: using the intermediate deep learning network as the original deep learning network, the previous steps of acquiring data samples, encoding, cross-collision, response output, and optimization are repeated until debugging is complete. This iterative process continuously adjusts the network parameters, gradually improving the network's predictive ability. When certain stopping conditions are met (such as reaching a preset number of iterations or the loss function value being less than a certain threshold), the target deep learning network is obtained. Finally, based on the target knowledge vector encoding module and target knowledge cross-collision module in the target deep learning network, an optimized initial AI knowledge embedding network is obtained.

[0070] It is evident that by loading the associated conversational interaction knowledge information set into the initial AI knowledge embedding network for knowledge vector encoding and cross-collision, the powerful feature extraction capabilities of deep learning networks are leveraged to process knowledge information more accurately. During the debugging process, based on computing resource conversation data samples and prior conversation response training annotations, the original deep learning network was continuously optimized, improving the network's response accuracy to different types of computing resource conversations. The knowledge vector encoding module employs specific algorithms, such as CNN-based encoding, which effectively captures semantic information in the conversation. The knowledge cross-collision module further integrates information from different knowledge encodings. The entire debugging loop ensures that the network gradually converges to the optimal state, and the resulting target AI knowledge embedding network operates more efficiently and accurately in computing resource conversation response processing, enhancing the overall system's ability to handle computing resource conversations.

[0071] In some preferred embodiments, the step of identifying the target session interaction knowledge in the session data of the computing power resources to be processed and obtaining the target session interaction knowledge information set includes: mining the session interaction semantic vector of the session data of the computing power resources to be processed; performing feature preprocessing based on the session interaction semantic vector to obtain a session interaction semantic preprocessing vector; decoding the session interaction semantic preprocessing vector for session requirements to obtain a session requirement decoding vector; and collecting information based on the session requirement decoding vector to obtain the target session interaction knowledge information set.

[0072] It is understandable that, in order to mine the semantic vectors of conversational interactions in the computational resource conversation data to be processed, a method combining word embedding techniques and deep learning architectures can be adopted. For example, a pre-trained Bidirectional Encoder Representation (BERT) model can be used. The BERT model is pre-trained on a large-scale text corpus and can capture rich semantic information.

[0073] For example, the session data of computing resources to be processed can be represented as a text sequence (T = {t_1, t_2, cdots, t_n}), where (t_i) represents the (i)th word or token in the session data. This text sequence is then input into the BERT model. The BERT model contains a multi-layer Transformer structure. In the Transformer structure, for each input word token (t_i), its relationship with other word tokens is calculated using a multi-head attention mechanism.

[0074] The calculation formula for the multi-head attention mechanism is as follows: First, for the attention calculation of a single head, (Attention(Q, K, V)=softmax(frac{QK^T}{sqrt{d_k}})V), where (Q), (K), and (V) are the query vector, key vector, and value vector, respectively, which are obtained by linearly transforming the input word tag vector, and (d_k) is the dimension of the key vector.

[0075] In a multi-head attention mechanism, the attention results from multiple heads are concatenated, and then a linear transformation is performed to obtain the final result. For example, using an attention mechanism with (h) heads, for each head (i), (Attention_i(Q, K, V)) is calculated, and the final multi-head attention result (MultiHead(Q, K, V) = Concat(Attention_1(Q, K, V), cdots, Attention_h(Q, K, V)) W^O) is the output linear transformation weight matrix.

[0076] After multiple layers of processing by the BERT model, the vector output by the last layer can be used as the session interaction semantic vector (\(\vec{s}\)). This vector contains the semantic information of the entire session data, and its dimension depends on the settings of the BERT model, for example, it is 768 dimensions.

[0077] Then, after obtaining the session interaction semantic vector (\(\vec{s}\)), feature preprocessing operations are performed. This may include normalization, dimensionality reduction, etc.

[0078] For normalization, the min-max normalization method can be used. For example, for the session interaction semantic vector \(\vec{s}=(s_1, s_2, \cdots, s_d)\) (\(d\) is the vector dimension, such as \(d = 768\)), the min-max normalization formula is \(s_i'=\frac{s_i - min(s)}{max(s)-min(s)}\), where \(s_i'\) is the \(i\)-th element after normalization, and \(min(s)\) and \(max(s)\) are the minimum and maximum values in the vector \(\vec{s}\) respectively. By normalization, the values of the vector elements are mapped to the interval \([0, 1]\), which helps to improve the stability of subsequent processing.

[0079] For dimensionality reduction, the principal component analysis (PCA) method can be used. For example, the covariance matrix of the original session interaction semantic vector \(\vec{s}\) is \(\Sigma\), calculate the eigenvalues \((\lambda_1, \lambda_2, \cdots, \lambda_d)\) and the corresponding eigenvectors \((\vec{u}_1, \vec{u}_2, \cdots, \vec{u}_d)\) of \(\Sigma\). Sort the eigenvectors according to the eigenvalue size, and select the first \(k\) eigenvectors (\(k < d\), for example, \(k = 128\)) to construct the projection matrix \(P = [\vec{u}_1, \vec{u}_2, \cdots, \vec{u}_k]\). The session interaction semantic preprocessing vector \(\vec{s}' = P^T\vec{s}\), reducing the original \(d\)-dimensional vector to \(k\) dimensions.

[0080] In this embodiment, the session requirement decoding method can adopt a method based on a recurrent neural network (RNN) for session requirement decoding. Specifically, a long short-term memory network (LSTM) is used.

[0081] The elements of the session interaction semantic preprocessing vector \(\vec{s}'\) are sequentially input into the LSTM network. The core unit of the LSTM network includes an input gate (\(i_t\)), a forget gate (\(f_t\)), an output gate (\(o_t\)) and a cell state (\(c_t\)).

[0082] The calculation formula of the input gate is:

[0083] (i_t=sigma(W_{xi}x_t+W_{hi}h_{t-1}+b_i)), where (x_t) are the elements of the input vector (i.e., the elements in (vec{s}'), (W_{xi}) and (W_{hi}) are the weight matrices, (h_{t-1}) is the hidden state at the previous time step, (b_i) is the bias vector, and (sigma) is the sigmoid function.

[0084] The formula for calculating the forgetting gate is:

[0085] (f_t=sigma(W_{xf}x_t+W_{hf}h_{t-1}+b_f)).

[0086] The cell state update formula is (c_t=f_tc_{t-1}+i_ttanh(W_{xc}x_t+W_{hc}h_{t-1}+b_c)).

[0087] The formula for calculating the output gate is:

[0088] (o_t=sigma(W_{xo}x_t+W_{ho}h_{t-1}+b_o)), the final hidden state (h_t=o_ttanh(c_t)).

[0089] After processing by the LSTM network, the hidden state (h_T) at the last time step (e.g., the input sequence length is (T)) can be used as the session requirement decoding vector (vec{d}).

[0090] Finally, information is collected based on the session requirement decoding vector to obtain the target session interaction knowledge information set: information is collected based on the session requirement decoding vector (vec{d}). A rule-based information extraction method combined with a predefined knowledge base can be used.

[0091] For example, the knowledge base contains various knowledge templates related to computing resources, such as computing resource demand templates and computing resource fault templates. For the session demand decoding vector (vec{d}), its similarity to each knowledge template in the knowledge base is calculated. The cosine similarity calculation method can be used; for the vector (vec{d}) and the knowledge template vector (vec{k}) in the knowledge base, the cosine similarity is calculated as (cos(vec{d}, vec{k})).

[0092] =frac{vec{d}cdotvec{k}}{vertvec{d}vertvertvec{k}vert}).

[0093] Based on the calculated similarity, the knowledge content corresponding to the knowledge template with the highest similarity is selected as the target session interaction knowledge information set. For example, if the similarity is highest with the computing resource requirement template, then the target session interaction knowledge information set will contain relevant knowledge content about computing resource requirements, such as the type of requirement (whether it is computing power requirement, storage requirement, or network bandwidth requirement, etc.) and the scale of requirement (specific numerical requirements, etc.).

[0094] This design, firstly, effectively captures the semantic information of the computing resource session data by mining the semantic vectors of conversational interactions and utilizing multi-head attention mechanisms such as those in the BERT model. Normalization and dimensionality reduction operations in feature preprocessing improve data stability and reduce computational load. LSTM is used for session requirement decoding; its internal gating mechanism effectively handles long-distance dependencies in sequence data, accurately transforming the preprocessed semantic vectors into session requirement decoding vectors. Finally, information is collected based on these decoding vectors. By calculating similarity with knowledge templates in the knowledge base, the target session interaction knowledge information set can be accurately determined. This helps improve the accuracy of the entire computing resource session response processing system's understanding of session content, thereby enhancing the system's response quality and efficiency.

[0095] In some other preferred embodiments, determining the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector from the associated session interaction knowledge vectors corresponding to each computing resource business keyword based on the correlation feature values ​​of the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing resource business keyword includes: determining the feature commonality score of the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing resource business keyword to obtain each correlation feature value; determining the target correlation feature value from the various correlation feature values, and using the associated session interaction knowledge vector corresponding to the target correlation feature value as the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector.

[0096] In practical applications, the following is a detailed description of the technical solution described in this embodiment:

[0097] 1. Determine the common feature scores of the current session interaction knowledge vector and related session interaction knowledge vectors to obtain relevance feature values.

[0098] Selection of Feature Commonality Score Calculation Method: To determine the feature commonality score (i.e., relevance feature value) between the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing resource business keyword, various methods can be used. One effective method is a combination of cosine similarity calculation and semantic distance measurement based on the Vector Space Model (VSM).

[0099] Cosine similarity calculation: Let the current session interaction knowledge vector be (vec{v}_{current}), and its dimension be (d) (e.g. (d=300)). The associated session interaction knowledge vector corresponding to a certain computing power resource business keyword is (vec{v}_{associated}).

[0100] The formula for cosine similarity is:

[0101] (cosine(vec{v}_{current}, vec{v}_{associated})=

[0102] frac{vec{v}_{current}cdotvec{v}_{associated}}{vertvec{v}_{current}vertvertvec{v}_{associated}vert});

[0103] in:

[0104] (vec{v}_{current}cdotvec{v}_{associated}

[0105] =sum_{i=1}^{d}v_{current,i}v_{associated,i}),(vertvec{v}_{current}vert=sqrt{sum_{i=1}^{d}v_{current,i}^2}),(vertvec{v}_{associated}vert=sqrt{sum_{i=1}^{d}v_{associated,i}^2}). This cosine similarity value reflects the degree of similarity between two vectors in a direction, and the value ranges between ([-1, 1]). The closer the value is to (1), the more similar they are.

[0106] A supplement to semantic distance metrics: Cosine similarity alone may not fully and accurately reflect feature commonalities. Therefore, semantic distance metrics are introduced. For example, semantic knowledge bases such as WordNet are used to calculate the semantic distance between words. For instance, the words in the current session interaction knowledge vector and the associated session interaction knowledge vector are (w_{1}, w_{2}, cdots) and (w_{1}', w_{2}', cdots), respectively. For each pair of corresponding words (if they exist), their semantic distance (d(w_i, w_i')) is determined by querying WordNet.

[0107] An exemplary comprehensive calculation method is to perform a weighted summation of cosine similarity and semantic distance. Let the weights be (w_{cos}) and (w_{sem}) (e.g., (w_{cos} = 0.7), (w_{sem} = 0.3)), then the feature commonality score (relevance feature value) (r = w_{cos} * cosine(vec{v}_{current}, vec{v}_{associated}) + w_{sem} * frac{1}{1 + sum_{i}d(w_i, w_i')}). In this formula, (frac{1}{1 + sum_{i}d(w_i, w_i')}) is a normalization process for the semantic distance, ensuring that the influence of semantic distance is integrated with cosine similarity on the same order of magnitude.

[0108] II. Determine the target relevance feature value from various relevance feature values ​​and determine the target associated conversation interaction knowledge vector.

[0109] Methods for determining target relevance feature values: After obtaining the various relevance feature values, it is necessary to determine the target relevance feature value from these values. A common method is to set a threshold (t) (e.g., (t = 0.6)) and filter out relevance feature values ​​that are greater than or equal to this threshold.

[0110] If multiple relevance feature values ​​satisfy this condition, further selection can be made based on other rules. For example, the largest relevance feature value can be selected as the target relevance feature value. This is because the largest relevance feature value indicates that the current session interaction knowledge vector and the corresponding associated session interaction knowledge vector are most similar in feature commonality, and are most likely to contain knowledge content related to the current session.

[0111] Determine the target associated session interaction knowledge vector: Once the target relevance feature value is determined, the associated session interaction knowledge vector corresponding to this target relevance feature value is used as the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector. For example, if the relevance feature value (r^*) corresponding to a certain associated session interaction knowledge vector (vec{v}_{associated}^*) is determined as the target relevance feature value, then (vec{v}_{associated}^*) is the target associated session interaction knowledge vector. This target associated session interaction knowledge vector will be used in subsequent steps to determine the session interaction business topic keywords, and thus determine the computing resource session response result corresponding to the computing resource session data to be processed.

[0112] The above embodiments, by employing cosine similarity combined with semantic distance metric to determine relevance feature values, can more comprehensively evaluate the commonalities in features between the current session interaction knowledge vector and related session interaction knowledge vectors. Cosine similarity measures directional similarity from the perspective of vector space, while semantic distance metric supplements the consideration of differences at the lexical semantic level. The relevance feature value obtained by weighted summation of the two more accurately reflects the actual relevance. When determining the target relevance feature value, by setting thresholds and further filtering rules, the most relevant vector can be accurately selected from multiple related session interaction knowledge vectors as the target related session interaction knowledge vector. This helps improve the accuracy of the entire computing power resource session response processing system, making the final determined session response result more in line with the actual needs of the session, thereby improving the system's performance and effectiveness.

[0113] In some examples, the computing power resource business keywords include computing power scheduling business keywords and cloud computing power mode update keywords. The step of determining the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector from the associated session interaction knowledge vectors corresponding to each computing power resource business keyword, based on the correlation feature values ​​of the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing power resource business keyword, includes: determining the feature commonality score between the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to the computing power scheduling business keyword to obtain a computing power scheduling correlation feature value; determining the feature commonality score between the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to the cloud computing power mode update keyword to obtain a mode update correlation feature value; and determining the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector from the associated session interaction knowledge vectors corresponding to the computing power scheduling business keyword and the associated session interaction knowledge vectors corresponding to the cloud computing power mode update keyword, based on the difference variable between the computing power scheduling correlation feature value and the mode update correlation feature value.

[0114] Further, determining the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector from the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the current session interaction knowledge vector based on the difference variable between the associated session interaction knowledge vector corresponding to ...

[0115] Similar to the previous method for determining relevance feature values, it is first necessary to determine the feature commonality score of the current session interaction knowledge vector and the associated session interaction knowledge vector corresponding to the computing power scheduling business keywords, so as to obtain the computing power scheduling relevance feature value.

[0116] For example, a method based on the Vector Space Model (VSM) combined with semantic analysis can be used. Let the knowledge vector of the current session interaction be (vec{v}_{current}), and the knowledge vector of the associated session interaction corresponding to the computing power scheduling business keyword be (vec{v}_{scheduling}).

[0117] Cosine similarity calculation based on the vector space model: (cosine(vec{v}_{current},vec{v}_{scheduling})

[0118] =frac{vec{v}_{current}cdotvec{v}_{scheduling}}{vertvec{v}_{current}vertvertvec{v}_{scheduling}vert}), where:

[0119] (vec{v}_{current}cdotvec{v}_{scheduling}

[0120] =sum_{i=1}^{d}v_{current,i}v_{scheduling,i}),(vertvec{v}_{current}vert=sqrt{sum_{i=1}^{d}v_{current,i}^2}),(vertvec{v}_{scheduling}vert=sqrt{sum_{i=1}^{d}v_{scheduling,i}^2}), where (d) is the vector dimension, for example (d=300).

[0121] Semantic analysis is performed simultaneously. For example, if the words in the two vectors are (w_{1}, w_{2}, cdots) and (w_{1}', w_{2}', cdots), for each pair of corresponding words (if they exist), their semantic distance (d(w_i, w_i')) is determined by querying a semantic knowledge base (such as WordNet). A comprehensive calculation method is to obtain the computational scheduling relevance feature value (r_{scheduling}) by weighted summing of cosine similarity and semantic distance. Let the weights be (w_{cos} = 0.7) and (w_{sem} = 0.3), then (r_{scheduling} = w_{cos} * cosine(vec{v}_{current}, vec{v}_{scheduling}) + w_{sem} * frac{1}{1 + sum_{i}d(w_i, w_i')}).

[0122] Determine the relevance feature value (mode update relevance feature value) to the cloud computing power mode update keyword: Similarly, determine the feature commonality score between the current session interaction knowledge vector and the associated session interaction knowledge vector corresponding to the cloud computing power mode update keyword to obtain the mode update relevance feature value.

[0123] Let the knowledge vector of the current session interaction be (vec{v}_{current}), and the knowledge vector of the associated session interaction corresponding to the cloud computing power mode update keyword be (vec{v}_{update}).

[0124] First, calculate the cosine similarity (cosine(vec{v}_{current},vec{v}_{update})=frac{vec{v}_{current}cdotvec{v}_{update}}{vertvec{v}_{current}vertvertve c{v}_{update}vert}).

[0125] Then, semantic analysis is performed to determine the semantic distance between words (d(w_i, w_i')). The pattern update relevance feature value (r_{update}) is obtained by weighted summation as described above, i.e. (r_{update}=w_{cos}*cosine(vec{v}_{current},vec{v}_{update})+w_{sem}*frac{1}{1+sum_{i}d(w_i,w_i')}).

[0126] II. Determining the target-related conversational knowledge vector based on distinguishing variables

[0127] Identify the distinguishing variable: The distinguishing variable reflects the degree of difference between the computational scheduling-related feature values ​​and the pattern update-related feature values. Calculate the distinguishing variable (Delta = vertr_{scheduling} - r_{update}vert).

[0128] When the differential variable is greater than the set variable value: The set variable value is a pre-determined threshold, for example, the set variable value is (0.2). When (Delta>0.2), a comparative analysis is performed on (r_{scheduling}) and (r_{update}).

[0129] If (r_{scheduling}>r_{update}), this means that the current session interaction knowledge vector is closer in feature commonality to the associated session interaction knowledge vector corresponding to the computing power scheduling business keyword. Therefore, the associated session interaction knowledge vector corresponding to the computing power scheduling business keyword is taken as the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector.

[0130] Conversely, if (r_{update}>r_{scheduling}), then the associated session interaction knowledge vector corresponding to the cloud computing power mode update keyword will be used as the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector.

[0131] Handling when the distinguishing variable is not greater than the set variable value: When (Deltaleq0.2), it indicates that the current session interaction knowledge vector, the associated session interaction knowledge vector corresponding to the computing power scheduling business keyword, and the associated session interaction knowledge vector corresponding to the cloud computing power mode update keyword do not differ significantly in terms of common features. In this case, the general business theme keyword is used as the session interaction business theme keyword of the target session interaction knowledge information set. The general business theme keyword is a predefined keyword with a broad coverage and related to various computing power resource businesses, such as "computing power resource basic management". This means that the current session interaction knowledge vector may not be clearly attributed to the computing power scheduling business or the cloud computing power mode update business, so the general business theme keyword is used to process subsequent session responses.

[0132] It is evident that by separately determining the relevance feature values ​​corresponding to computing power scheduling business keywords and cloud computing power mode update keywords, and employing a comprehensive vector space model and semantic analysis method, the correlation between the current session interaction knowledge vector and the knowledge vectors related to different business keywords can be accurately assessed. The target associated session interaction knowledge vector is determined based on the distinguishing variable between the computing power scheduling relevance feature values ​​and the mode update relevance feature values. Setting a variable value as a distinguishing boundary allows for reasonable decision-making under varying relevance feature value differences. When the difference is large, the most relevant session interaction knowledge vector corresponding to the business keyword can be accurately selected based on the specific magnitude of the difference; when the difference is small, general business topic keywords are used, avoiding misjudgment and improving the accuracy and flexibility of the entire computing power resource session response processing system, thereby enhancing the system's ability to handle different types of sessions.

[0133] In an optional embodiment, the target session interaction knowledge information set includes at least two; determining the computing resource session response result corresponding to the computing resource session data to be processed based on the session interaction business topic keywords of the target session interaction knowledge information set includes: determining the heat evaluation coefficients of the at least two target session interaction knowledge information sets with the computing resource session data to be processed, and taking the session interaction business topic keywords of the target session interaction knowledge information set with the largest heat evaluation coefficient as the computing resource session response result corresponding to the computing resource session data to be processed.

[0134] In this embodiment, the thermal evaluation coefficient is an indicator used to measure the degree of correlation between the target session interaction knowledge information set and the computing resource session data to be processed. By calculating this coefficient, it is possible to determine which target session interaction knowledge information set is most critical in the current session context, thereby providing a basis for determining the correct computing resource session response result.

[0135] For each target session interaction knowledge information set, a method based on the term frequency-inverse document frequency (TF-IDF) algorithm combined with semantic similarity is used to calculate its thermal evaluation coefficient with the session data of the computing power resources to be processed.

[0136] First, we perform word frequency statistics on the computing resource session data to be processed. Let the computing resource session data to be processed be (D), and let the set of words contained therein be (V={v_1,v_2,cdots,v_n}). For each word (v_i), its word frequency (tf(v_i,D)) in (D) is defined as the number of times that word appears in (D).

[0137] Then, calculate the inverse document frequency (idf(v_i)). For example, if there is a corpus (C = {D_1, D_2, cdots, D_m}) containing multiple documents (here, other historical computing resource session data can be regarded as documents), and the number of documents containing the word (v_i) is (df(v_i)), then (idf(v_i) = logfrac{m}{df(v_i)+1}). Here, (1) is added to avoid the case where the denominator is (0).

[0138] For a target session interaction knowledge information set (K_j) ((j = 1, 2, cdots) representing different target session interaction knowledge information sets), the vocabulary set it contains is (V_j = {v_{j1}, v_{j2}, cdots, v_{jn_j}}). Calculate the TF-IDF-based similarity (s_{TF-IDF}(K_j, D)) between (K_j) and (D):

[0139] First, calculate the TF-IDF value of each word (v_{ji}) in (K_j) in (D) (tf-idf(v_{ji},D)=tf(v_{ji},D)*idf(v_{ji})).

[0140] Then, (s_{TF-IDF}(K_j,D)=sum_{v_{ji}inV_jcapV}tf-idf(v_{ji},D)), this value reflects the similarity between (K_j) and (D) from the perspective of term frequency-inverse document frequency.

[0141] Simultaneously, semantic similarity is calculated. A pre-trained word vector model (such as Word2Vec) is used to obtain the word vectors of the words. For words (v_{ji}) in (K_j) and words (v_i) in (D), if there are the same words, their semantic similarity is (1); if they are different words, let their word vectors be (vec{w}_{ji}) and (vec{w}_{i}) respectively, then their semantic similarity is (s_{sem}(v_{ji},v_i)=frac{vec{w}_{ji}cdotvec{w}_{i}}{vertvec{w}_{ji}vertvertvec{w}_{i}vert}).

[0142] The thermal evaluation coefficient (h_j) is calculated by combining TF-IDF similarity and semantic similarity. Let the weights be (w_{TF-IDF}=0.6) and (w_{sem}=0.4), then (h_j=w_{TF-IDF}*s_{TF-IDF}(K_j,D)+w_{sem}*frac{1}{n_j}sum_{v_{ji}inV_jcapV}s_{sem}(v_{ji},vi)). Here, (frac{1}{n_j}sum_{v_{ji}inV_jcapV}s_{sem}(v_{ji},vi)) is the average of the semantic similarities between the shared words in (K_j) and (D).

[0143] After calculating the thermal evaluation coefficient (h_j) corresponding to each target session interaction knowledge information set, the magnitudes of these coefficients are compared. For example, there are three target session interaction knowledge information sets (K_1), (K_2), and (K_3), and their thermal evaluation coefficients are calculated to be (h_1 = 0.45), (h_2 = 0.52), and (h_3 = 0.48), respectively.

[0144] Since (h_2) is the largest, the session interaction business topic keyword of the target session interaction knowledge information set (K_2) is used as the computing resource session response result corresponding to the computing resource session data to be processed. This session interaction business topic keyword will be used to query the knowledge base or generate response content. If querying the knowledge base, the corresponding response information is found in the knowledge base using this topic keyword as an index; if generating response content, appropriate response content is generated based on the knowledge related to this topic keyword, according to a predefined template or rule.

[0145] Therefore, by calculating the heat map evaluation coefficient between the target session interaction knowledge information set and the session data to be processed, a method combining the term frequency-inverse document frequency (TF-IDF) algorithm and semantic similarity is achieved. The TF-IDF algorithm measures the importance of words in terms of their frequency and rarity in documents, thus reflecting the degree of association between the target session interaction knowledge information set and the session data to be processed at the lexical level. Semantic similarity further supplements the evaluation of association from the perspective of lexical semantics. The heat map evaluation coefficient obtained by weighted summation of the two methods more accurately reflects the actual closeness of association between the target session interaction knowledge information set and the session data to be processed. Selecting the session interaction business topic keywords of the target session interaction knowledge information set with the largest heat map evaluation coefficient as the response result improves the accuracy of determining the response result of the computing resource session, enabling the system to make more accurate and appropriate responses to different session situations, thereby improving the performance of the entire session response processing system.

[0146] Based on steps 202-208, the method further includes: loading the target session interaction knowledge information set and the associated session interaction knowledge vectors corresponding to each computing resource business keyword into a bidirectional long short-term memory network; mining the current session interaction knowledge vector corresponding to the target session interaction knowledge information set through the knowledge vector encoding module in the bidirectional long short-term memory network, and determining the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector from the associated session interaction knowledge vectors corresponding to each computing resource business keyword based on the correlation feature values ​​of the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing resource business keyword; and using the computing resource business keyword corresponding to the target associated session interaction knowledge vector as the session interaction business topic keyword of the corresponding target session interaction knowledge information set.

[0147] Furthermore, the debugging method for the bidirectional long short-term memory network includes: acquiring a sample of the session interaction knowledge information set and the corresponding scenario session response training annotations; loading the sample of the session interaction knowledge information set and the associated session interaction knowledge vectors corresponding to each computing power resource service keyword into the original bidirectional long short-term memory network for session keyword parsing to obtain the session interaction service prediction keywords corresponding to the sample of the session interaction knowledge information set; optimizing the original bidirectional long short-term memory network based on the session interaction service prediction keywords and the scenario session response training annotations to obtain an optimized bidirectional long short-term memory network; using the optimized bidirectional long short-term memory network as the original bidirectional long short-term memory network, and jumping back to the step of acquiring the sample of the session interaction knowledge information set and the corresponding scenario session response training annotations, repeating the process until the debugging process is completed to obtain the bidirectional long short-term memory network.

[0148] Based on steps 202-208, the target session interaction knowledge information set and the associated session interaction knowledge vectors corresponding to the business keywords of each computing resource are loaded into a bidirectional long short-term memory network. A bidirectional long short-term memory network is a special type of recurrent neural network (RNN) that can handle long-distance dependencies in sequential data, and due to its bidirectional structure, it can consider both the forward and backward information of the sequence simultaneously.

[0149] The knowledge vector encoding module in the bidirectional long short-term memory network is responsible for mining the current session interaction knowledge vector corresponding to the target session interaction knowledge information set.

[0150] Let the target session interaction knowledge information set be represented as (K = {k_1, k_2, cdots, k_n}), and input it sequentially into a bidirectional Long Short-Term Memory (LSTM) network. The basic unit of the bidirectional LSTM network is the Long Short-Term Memory unit (LSTM unit). For forward propagation, the LSTM unit is calculated as follows:

[0151] The forget gate (f_t = sigma(W_{xf}x_t + W_{hf}h_{t-1} + b_f)) is used, where (x_t) are the input elements (elements in the knowledge information set), (W_{xf}) and (W_{hf}) are the weight matrices, (h_{t-1}) is the hidden state at the previous time step, (b_f) is the bias vector, and (sigma) is the sigmoid function.

[0152] Input gate: (i_t=sigma(W_{xi}x_t+W_{hi}h_{t-1}+b_i)).

[0153] Cell state:

[0154] (c_t=f_tc_{t-1}+i_ttanh(W_{xc}x_t+W_{hc}h_{t-1}+b_c)).

[0155] Output gate:

[0156] (o_t=sigma(W_{xo}x_t+W_{ho}h_{t-1}+b_o)), the final hidden state (h_t=o_ttanh(c_t)).

[0157] For backpropagation, the computation process is similar, only the input order is reversed. Finally, the hidden states obtained from forward and backpropagation are combined (e.g., concatenated) to obtain the current session interaction knowledge vector (vec{v}_{current}).

[0158] Then, based on the correlation feature values ​​between the current session interaction knowledge vector (vec{v}_{current}) and the associated session interaction knowledge vectors corresponding to each computing power resource business keyword, the target associated session interaction knowledge vector is determined from these associated session interaction knowledge vectors.

[0159] Cosine similarity can be used to calculate the relevance feature value. Let (vec{v}_{associated}) be the knowledge vector of the associated conversation interaction corresponding to a certain computing resource business keyword. Then the cosine similarity (cosine(vec{v}_{current},vec{v}_{associated})=frac{vec{v}_{current}cdotvec{v}_{associated}}{vertvec{v}_{current}vertver tvec{v}_{associated}vert});

[0160] in:

[0161] (vec{v}_{current}cdotvec{v}_{associated}

[0162] =sum_{i=1}^{d}v_{current,i}v_{associated,i}),(vertvec{v}_{current}vert=sqrt{sum_{i=1}^{d}v_{current,i}^2}),(vertvec{v}_{associated}vert=sqrt{sum_{i=1}^{d}v_{associated,i}^2}), where (d) is the vector dimension, for example (d=300). By calculating the cosine similarity with each associated session interaction knowledge vector, the associated session interaction knowledge vector corresponding to the maximum cosine similarity is found as the target associated session interaction knowledge vector. The computing power resource business keywords corresponding to the target associated session interaction knowledge vector are used as the session interaction business topic keywords of the corresponding target session interaction knowledge information set.

[0163] Furthermore, the debugging method for bidirectional long short-term memory networks is as follows.

[0164] 1) Obtaining Data Samples and Training Annotations: First, obtain the conversation interaction knowledge information set samples and the corresponding scenario conversation response training annotations. The conversation interaction knowledge information set samples are representative samples extracted from a large amount of computing resource conversation data, while the scenario conversation response training annotations are the annotation information of the correct response results corresponding to these samples.

[0165] 2) Obtain predicted keywords by parsing conversation keywords: Load the sample of conversation interaction knowledge information set and the associated conversation interaction knowledge vectors corresponding to the business keywords of each computing resource into the original bidirectional long short-term memory network for conversation keyword parsing.

[0166] Following the computational method of the bidirectional long short-term memory network described above, the input conversational interaction knowledge information set sample is processed to finally obtain the conversational interaction business prediction keywords corresponding to the conversational interaction knowledge information set sample.

[0167] 3) Optimize the network based on predicted keywords and training annotations: Optimize the original bidirectional long short-term memory network based on predicted keywords for conversational interaction services and training annotations for scenario conversation responses.

[0168] The cross-entropy loss function can be used to measure the difference between predicted keywords and correct keywords. Let the probability distribution of predicted conversational interaction business keywords be (y^{pred}={y_1^{pred},y_2^{pred},cdots,y_m^{pred}})(where (m) is the number of possible keyword types), and the probability distribution of correct conversational interaction business keywords be (y^{true}={y_1^{true},y_2^{true},cdots,y_m^{true}})(usually one-hot encoding is used, that is, the correct type is (1), and the others are (0)), then the cross-entropy loss function is (L=-sum_{i=1}^{m}y_i^{true}log(y_i^{pred})).

[0169] By using the backpropagation algorithm, the weight parameters in the bidirectional long short-term memory network are updated based on the gradient calculated from the loss function, thus obtaining an optimized bidirectional long short-term memory network.

[0170] 4) Iterative debugging to obtain the final network: Using the optimized bidirectional long short-term memory (BSSM) network as the original BSSM network, the previous steps of obtaining samples, parsing, and optimizing are repeated until the debugging process is completed, resulting in the BSSM network. This iterative process continuously adjusts the network parameters, gradually improving the network's ability to predict session interaction business keywords. The final BSSM network can more accurately handle the relationship between the target session interaction knowledge information set and the associated session interaction knowledge vector, thereby improving the performance of the entire computing resource session response processing system.

[0171] This design, by loading the target session interaction knowledge information set and associated session interaction knowledge vectors into a bidirectional long short-term memory (LSTM) network, leverages its bidirectional processing capability for sequence data to more comprehensively mine the current session interaction knowledge vectors. During the mining process, the computational method of the LSTM network effectively captures long-distance dependencies in the knowledge information, improving the accuracy of knowledge vector mining. When determining the target associated session interaction knowledge vectors, methods such as cosine similarity can accurately find the most relevant vectors, thereby determining accurate session interaction business topic keywords. The debugging method of the LSTM network, based on session interaction knowledge information set samples and scenario session response training annotations, continuously optimizes the network through cross-entropy loss function and backpropagation algorithm, ensuring that the network's prediction accuracy gradually improves, thereby enhancing the efficiency and accuracy of the entire system in processing sessions using computing resources.

[0172] It is worth mentioning that the unique application and innovation of Bi-LSTM network in the above technical solution are explained as follows.

[0173] I. Unique Applications of Bidirectional Long Short-Term Memory Networks in Technical Solutions

[0174] (I) Handling knowledge vector relationships

[0175] Mining Current Session Interaction Knowledge Vectors: In the above technical solution, the knowledge vector encoding module of a Bidirectional Long Short-Term Memory (Bi-LSTM) network is used to mine the current session interaction knowledge vectors corresponding to the target session interaction knowledge information set. The unique structure of Bi-LSTM allows it to simultaneously consider both the forward and backward information of the input sequence (target session interaction knowledge information set).

[0176] For example, when the target session interaction knowledge set contains multiple elements, traditional unidirectional networks can only process the relationships between these elements in one direction (e.g., from front to back), potentially losing some reverse semantic information. Bi-LSTM, however, processes these relationships simultaneously using forward and reverse LSTM units, enabling a more comprehensive capture of dependencies between elements. Taking an exemplary target session interaction knowledge set containing three elements (k_1, k_2, k_3) as an example, the forward LSTM unit calculates a hidden state based on (k_1), then combines it with (k_2) to calculate the next hidden state, and so on. The reverse LSTM unit starts from (k_3) and calculates the hidden states in reverse. Finally, the forward and reverse hidden states are combined (e.g., concatenated) to obtain the current session interaction knowledge vector. This approach can better uncover the potential semantic and logical relationships within the knowledge set, resulting in a more accurate current session interaction knowledge vector.

[0177] Determining the target-related session interaction knowledge vector: The output of Bi-LSTM (the current session interaction knowledge vector) is used to determine the relevance feature values ​​of the related session interaction knowledge vectors corresponding to the business keywords of each computing resource. Because the knowledge vectors mined by Bi-LSTM contain richer information, the relationship between the target session interaction knowledge information set and the related session interaction knowledge vectors can be more accurately reflected when calculating the relevance feature values ​​(such as using cosine similarity).

[0178] For example, if the related session interaction knowledge vector and the current session interaction knowledge vector have complex semantic and logical connections (which may be positive, negative, or span multiple elements), the knowledge vector mined by Bi-LSTM can better capture this relationship, thereby improving the accuracy of determining the target related session interaction knowledge vector. By finding the related session interaction knowledge vector with the highest relevance feature value, its corresponding computing resource business keywords can be accurately identified as session interaction business topic keywords, which is crucial for determining the accurate computing resource session response results subsequently.

[0179] (II) Network debugging and optimization

[0180] Sample-based keyword parsing: In the debugging method of bidirectional long short-term memory networks, it is used to parse conversation keywords from samples of conversation interaction knowledge information. The associated conversation interaction knowledge vectors corresponding to the samples of conversation interaction knowledge information and the business keywords of each computing resource are loaded into the original bidirectional long short-term memory network.

[0181] Unlike general keyword parsing methods, Bi-LSTM takes into account long-distance dependencies and bidirectional semantic information in the samples during the parsing process. For example, for a long sample of conversational interaction knowledge information, there may be semantic relationships between early and late elements. Bi-LSTM can effectively capture such relationships through its bidirectional structure, thereby more accurately parsing the conversational interaction business prediction keywords.

[0182] Optimizing the network to improve accuracy: Based on the parsed session interaction business prediction keywords and scenario session response training annotations, the bidirectional long short-term memory network is optimized using the cross-entropy loss function and backpropagation algorithm. Since Bi-LSTM can better handle the semantic information in the samples during the initial parsing process, the network parameters can be adjusted more effectively during optimization.

[0183] For example, when calculating the cross-entropy loss function, if there is a deviation between the predicted keywords and the correct keywords, the Bi-LSTM structure helps to determine which parts of the sequence are misunderstood semantically. This allows for more targeted adjustment of weight parameters during backpropagation, improving the network's accuracy in predicting keywords for conversational interactions. This optimization process is continuously iterated, gradually improving the network's performance and ultimately resulting in a bidirectional long short-term memory network capable of accurately handling computationally relevant session tasks.

[0184] II. The novelty of the bidirectional long short-term memory network in the above technical solution

[0185] (I) Comparison with traditional technical methods

[0186] Limitations of Traditional Networks: In traditional techniques for processing knowledge vector relationships related to computing resource sessions, exemplary unidirectional neural networks or rule-based methods are often employed. Unidirectional neural networks (such as ordinary RNNs or LSTMs) can only capture information in one direction when processing sequential data, easily losing reverse semantic information. For example, when processing computing resource session data containing multiple sub-concepts, if earlier and later concepts have reverse logical relationships, unidirectional networks may fail to capture them effectively.

[0187] While rule-based methods can handle knowledge vector relationships according to predefined rules, they lack the ability to adaptively learn from data and struggle to handle complex and ever-changing computing resource session scenarios. For example, when new computing resource business keywords or new session patterns emerge, rule-based methods find it difficult to effectively expand and adapt.

[0188] The advantages of Bi-LSTM lie in its innovation: the application of bidirectional long short-term memory networks in the aforementioned technical solutions overcomes the limitations of traditional techniques. Through its bidirectional structure, it processes both forward and reverse information simultaneously, enabling it to better adapt to the complex semantic relationships within computational resource session data. For example, when determining the knowledge vectors of target-related session interactions, it can more accurately reflect the correlation between knowledge vectors in different directions, something that traditional unidirectional networks struggle to achieve.

[0189] Moreover, the application of Bi-LSTM in network debugging, by leveraging its ability to effectively capture long-distance dependencies, allows for more precise parameter adjustments to improve accuracy during network optimization. This approach of combining Bi-LSTM with specific computational resource session processing tasks (such as knowledge vector mining, target association vector determination, and network debugging) is not a conventional technique in this field and demonstrates originality.

[0190] (ii) Innovative combinations in specific technical solutions

[0191] Collaboration with other modules: In the above technical solution, Bi-LSTM collaborates with other modules (such as the knowledge vector encoding module and the module that determines the target associated session interaction knowledge vector based on relevance feature values) to form a complete computing resource session response processing system. This collaborative approach is unique. For example, the knowledge vector encoding module uses the current session interaction knowledge vector mined by Bi-LSTM for subsequent processing. The various modules depend on and interact with each other, jointly realizing the entire process from the target session interaction knowledge information set to determining the session interaction business topic keywords.

[0192] This specific combination of modules and collaborative working method is unprecedented in previous technologies in this field and is not a conventional technique that would be easily conceived by those skilled in the art. It embodies innovative thinking in solving the problem of processing session responses for computing resources, integrating Bi-LSTM into the architecture of the entire technical solution, thereby improving the overall performance and accuracy of the system.

[0193] Customization for Computing Resource Session Processing: The application of Bi-LSTM in this technical solution is customized for the specific domain of computing resource session processing. For example, when processing the relationship between computing resource business keywords and the target session interaction knowledge information set, the parameter adjustments and application methods of Bi-LSTM are designed to better adapt to the semantic characteristics and business logic of the computing resource domain. This domain-specific customization is not a conventional technique in this field, further demonstrating its innovation.

[0194] In some scalable embodiments, after using the computing resource business keywords corresponding to the target associated session interaction knowledge vector as the session interaction business topic keywords of the corresponding target session interaction knowledge information set, and determining the computing resource session response result corresponding to the computing resource session data to be processed based on the session interaction business topic keywords of the target session interaction knowledge information set, the method further includes: obtaining a consultation response event data sequence based on the computing resource session response result, wherein the consultation response event data sequence includes multiple consecutive sets of consultation response event data; obtaining a targeted push event data sequence based on the consultation response event data sequence, wherein the targeted push event data sequence includes multiple consecutive sets of targeted push event data; and based on the... The consultation response event data sequence is processed by obtaining a frequent consultation response item sequence through a first residual processing branch of the resource session analysis network, wherein the frequent consultation response item sequence includes multiple frequent consultation response items; based on the targeted push event data sequence, a targeted push frequent item sequence is obtained through a second residual processing branch of the resource session analysis network, wherein the targeted push frequent item sequence includes multiple targeted push frequent items; based on the frequent consultation response item sequence and the targeted push frequent item sequence, push label information corresponding to the consultation response event data is obtained through a push discrimination branch of the resource session analysis network; and targeted user profile information of the consultation response event data sequence is determined according to the push label information.

[0195] The above technical solution first obtains the consultation response event data sequence based on the computing power resource session response results.

[0196] The concept and composition of the consultation response event data sequence: After using the computing resource business keywords corresponding to the target-related session interaction knowledge vector as the session interaction business topic keywords, and determining the computing resource session response results corresponding to the computing resource session data to be processed based on these keywords, the consultation response event data sequence must first be obtained based on these response results. The consultation response event data sequence is an ordered set of data groups related to the consultation response, where each group of consultation response event data contains multifaceted information about a specific consultation response interaction.

[0197] For example, for a consultation regarding computing resource allocation, the response provides a specific allocation strategy. Related consultation-response event data may include the consultation timestamp, the consultant's identification (which could be a unique user ID or department ID), the specific content of the consultation (such as details of computing resource requirements for a specific task), the response content (such as recommended resource allocation schemes, related cost estimates, etc.), and some metadata about the consultation-response interaction (such as the interaction channel, whether through a web interface, mobile application, or internal system interface).

[0198] Then, the targeted push event data sequence is obtained from the consultation response event data sequence.

[0199] Method for obtaining targeted push event data sequences: Targeted push event data sequences are obtained based on consultation response event data sequences. These sequences also consist of multiple consecutive sets of data, each related to a specific targeted push operation.

[0200] To obtain targeted push event data sequences, rule-based mapping methods can be used. For example, for certain specific elements in the consultation response event data (such as consultant identification, specific keywords in the consultation content, etc.), predefined rules can be set to determine the corresponding targeted push content. For instance, if the consultation content contains the keyword "deep learning task computing power requirements," according to the predefined rules, the targeted push event data might include pushing links to technical documents related to deep learning computing power optimization, information on preferential packages offered by cloud computing power service providers for deep learning tasks, etc. This information related to targeted pushes, arranged chronologically or in relation to the consultation response events, forms the targeted push event data sequence.

[0201] Next, the frequent consultation response sequence is obtained through the first residual processing branch of the resource session analysis network.

[0202] Resource Session Analysis Network and First Residual Processing Branch: The Resource Session Analysis Network is a deep learning network specifically built for analyzing data related to computing resource sessions. The first residual processing branch processes the consultation-response event data sequence to obtain the sequence of frequent consultation-response items.

[0203] The residual processing branch adopts the idea of ​​the ResNet structure, which avoids the gradient vanishing or gradient exploding problems in deep neural networks during training by constructing residual blocks. For example, the first residual processing branch consists of multiple residual blocks, each containing two paths: one is the identity mapping path, and the other is the path through convolutional layers, batch normalization layers, and activation function layers (such as ReLU).

[0204] Let the sequence of consultation response event data input to the first residual processing branch be (X = {x_1, x_2, cdots, x_n}), where (x_i) represents the (i)th group of consultation response event data. For a single residual block, for example, if the input is (y), the output after the convolutional layer is (z = W*y + b) (where (W) is the convolutional kernel weight matrix and (b) is the bias vector), and the output after batch normalization is:

[0205] (z'=gamma*frac{z-mu}{sqrt{sigma^2+epsilon}}+beta)(where (gamma) and (beta) are learnable parameters, (mu) is the mean, (sigma^2) is the variance, and (epsilon) is a small constant to avoid a denominator of (0)), and the output after passing through the ReLU activation function is (z”=max(0,z')). Then (z”) is added to (y) of the identity mapping path to obtain the output of the residual block. Multiple residual blocks process the input data in sequence.

[0206] Obtaining the sequence of frequent consultation responses: After multiple layers of processing in the first residual processing branch, the output is used to obtain the sequence of frequent consultation responses. Frequent item mining algorithms, such as the Apriori algorithm, can be used.

[0207] For example, the output data after processing by the first residual processing branch is represented as (Y = {y_1, y_2, cdots, y_m}), which can be regarded as a transaction database, where each (y_i) is a transaction (containing multiple items, corresponding to different elements in the consultation response event data). For the Apriori algorithm, the minimum support threshold (s) is first determined (e.g., (s = 0.3)), which is the minimum proportion of an itemset appearing in all transactions.

[0208] The algorithm first scans all transactions, counts the occurrences of individual items, and finds frequent (1)-itemsets (L_1) that meet the minimum support threshold. Then, it generates candidate itemsets (C_2) from (L_1), scans the transaction database again to count the occurrences of itemsets in (C_2), and finds frequent (2)-itemsets (L_2) that meet the threshold. This process continues until no new frequent itemsets can be generated. The final sequence of frequent itemsets is the sequence of frequent itemsets in the consultation response, where each frequent item may be a specific consultant identity, frequently occurring keywords in the consultation content, or common response types, etc.

[0209] Furthermore, the sequence of frequently pushed items is obtained through the second residual processing branch of the resource session analysis network.

[0210] The operation of the second residual processing branch is similar to that of the first residual processing branch. In the resource session analysis network, the second residual processing branch is used to process the targeted push event data sequence to obtain the sequence of frequently pushed items. It also uses residual blocks of the residual network structure for processing.

[0211] Suppose the input sequence of targeted push event data is (A = {a_1, a_2, cdots, a_p}), which is processed by multiple residual blocks. The calculation method of each residual block is similar to that of the residual block in the first residual processing branch. For example, for input (b), after passing through the convolutional layer (W_1*b+b_1), batch normalization, activation function, and other operations, it is added to the identity mapping path to obtain the output. Multiple residual blocks process the input data sequentially.

[0212] Obtain the sequence of frequent items for targeted push notifications: Use a frequent item mining algorithm (such as the Apriori algorithm) similar to that used to obtain the sequence of frequent items for consultation responses to process the output data after processing the second residual processing branch.

[0213] For example, if the processed output data is (B = {b_1, b_2, cdots, b_q}), a minimum support threshold is set (e.g., (s' = 0.2)). By scanning the data and counting the occurrences of itemsets, frequent itemsets that meet the threshold are identified. These frequent itemsets form a sequence of frequent items for targeted push notifications. These frequent items may be content types that are frequently pushed to specific user groups, or notifications that are frequently received by a particular user group.

[0214] Next, push tag information is obtained based on the frequent item sequence of consultation response and the frequent item sequence of targeted push.

[0215] The working principle of the push discrimination branch: In the resource session analysis network, the push discrimination branch obtains the push tag information corresponding to the consultation response event data based on the frequent consultation response item sequence and the frequent targeted push item sequence. The push discrimination branch can use a classification model based on logistic regression.

[0216] Let the sequence of frequent consultation responses be (F = {f_1, f_2, cdots, f_r}), and the sequence of frequent targeted push notifications be (G = {g_1, g_2, cdots, g_s}). Combining and feature-engineering the elements of these two sequences yields a feature vector (vec{x}). For example, the frequency of occurrence of frequent items and the combination relationships between different frequent items can be converted into numerical features.

[0217] For a logistic regression model, for example, if the model parameters are (theta=(theta_0,theta_1,cdots,theta_t)), the basic formula for logistic regression is (h_{theta}(vec{x})=frac{1}{1+e^{-theta^Tvec{x}}});

[0218] Where (theta^Tvec{x}=theta_0+theta_1x_1+cdots+theta_tx_t). The formula calculates (h_{theta}(vec{x})) as a value between (0) and (1), representing the probability of belonging to a certain push label (e.g., whether it is a different type of push label such as "high value push" or "low value push").

[0219] Determine push label information: Based on the probability value calculated by the logistic regression model, determine the push label information by setting a threshold (e.g., (0.5)). If (h_{theta}(vec{x}) ≥ 0.5), the push label is determined to be of one category (e.g., "high-value push"); if (h_{theta}(vec{x}) < 0.5), it is determined to be of another category (e.g., "low-value push").

[0220] Finally, targeted user profile information for the consultation response event data sequence is determined based on the push notification tag information.

[0221] The construction of targeted user profile information is based on the following: push notification tags reflect certain characteristics of the consultation and response event data. Based on these tags, targeted user profile information for the consultation and response event data sequence can be determined. Targeted user profile information is a comprehensive description of the inquirer (or user group), including user characteristics related to computing resource consultation and response and targeted push notifications.

[0222] For example, if push notification information indicates that a user's inquiry response events are frequently marked as "high-value pushes," then the targeted user profile information may include characteristics such as that the user is a high-demand, high-value user, may have in-depth business needs for computing resources, and may be a potential long-term customer.

[0223] Cluster analysis can be used to cluster consultation and response event data with similar push notification tags, with each cluster representing a type of user. For example, using the K-Means clustering algorithm, let's assume the number of cluster centers is K = 3. First, randomly initialize (K) cluster centers (mu_1, mu_2, mu_3). Then, for each consultation and response event data, calculate its distance to each cluster center (e.g., Euclidean distance (d(x, mu) = sqrt{sum_{i=1}^{n}(x_i-mu_i)^2})) and assign it to the nearest cluster. Then recalculate the center of each cluster, repeating this process until the cluster centers no longer change significantly. Users in each cluster then possess similar targeted user profile information.

[0224] This design, firstly, systematically integrates various data related to computing resource session responses by constructing consultation response event data sequences and targeted push event data sequences, providing a comprehensive data foundation for subsequent analysis. Frequent item sequences are obtained using the residual processing branch in the resource session analysis network. The residual network structure ensures effectiveness in deep data processing, avoiding problems encountered in traditional deep network training. Frequent item mining algorithms such as Apriori are employed to uncover frequent patterns in the data. These frequent items are highly helpful in understanding common elements in consultation responses and targeted pushes. The push discrimination branch accurately determines push label information based on a logistic regression model. This classification method facilitates targeted categorization of consultation response events. Finally, targeted user profiles are constructed based on the push label information. Cluster analysis and other methods can better classify and understand users, providing valuable user insights for computing resource service providers and contributing to the development of more precise marketing strategies and service optimization plans.

[0225] Furthermore, Figure 2 This is a schematic diagram of the structure of a session response processing system 200 provided in this application. Figure 2 The session response processing system 200 shown includes a processor 210, which can call and run computer programs from memory to implement the methods in this application.

[0226] Optionally, such as Figure 2 As shown, the session response processing system 200 may further include a memory 230. The processor 210 can retrieve and run computer programs from the memory 230 to implement the methods described in this application.

[0227] The memory 230 can be a separate device independent of the processor 210, or it can be integrated into the processor 210.

[0228] Optionally, such as Figure 2 As shown, the session response processing system 200 may also include a transceiver 220, which the processor 210 can control to interact with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices.

[0229] Optionally, the session response processing system 200 can implement the corresponding processes of the storage engine or components (such as processing modules) in the storage engine or the device on which the storage engine is deployed in the various methods of this application. For the sake of brevity, these will not be elaborated here.

[0230] It should be understood that the processor in this application may be an integrated circuit chip with signal processing capabilities.

[0231] It is understood that the memory in this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory in the systems and methods described herein is intended to include, but is not limited to, suitable types of memory.

[0232] Based on the above, a readable storage medium is provided, on which a program or instructions are stored, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0233] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0234] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0235] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of protection of this application, and all of these forms are within the protection scope of this application.

Claims

1. A method for processing computing resource session responses applied to deep learning, characterized in that, The method is applied to a session response processing system, and the method includes: Obtain the knowledge information set of each associated session interaction corresponding to the set computing power resource business keywords; The knowledge vector encoding of each associated session interaction knowledge information set is performed to obtain the knowledge encoding of each associated session interaction. Cross-collision is performed on the various associated session interaction knowledge codes to obtain the associated session interaction knowledge vector corresponding to the set computing power resource business keyword; the various associated session interaction knowledge codes are combined to obtain the session interaction knowledge combination code; full connection processing is performed on the session interaction knowledge combination code to obtain the associated session interaction knowledge vector corresponding to the set computing power resource business keyword. Acquire the session data of computing resources to be processed, identify the target session interaction knowledge in the session data of computing resources to be processed, and obtain the target session interaction knowledge information set; Mining the current session interaction knowledge vector corresponding to the target session interaction knowledge information set, and obtaining the associated session interaction knowledge vector corresponding to each computing power resource business keyword. The associated session interaction knowledge vector is obtained by encoding each associated session interaction knowledge information set corresponding to the computing power resource business keyword into a knowledge vector, obtaining each associated session interaction knowledge code, and then cross-colliding the various associated session interaction knowledge codes. The computing power resource business keywords include computing power scheduling business keywords and cloud computing power mode update keywords. Based on the correlation feature values ​​between the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing power resource business keyword, the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector is determined from the associated session interaction knowledge vectors corresponding to each computing power resource business keyword. Then, the feature commonality score between the current session interaction knowledge vector and the associated session interaction knowledge vector corresponding to the computing power scheduling business keyword is determined to obtain the computing power scheduling correlation feature value; the feature commonality score between the current session interaction knowledge vector and the associated session interaction knowledge vector corresponding to the cloud computing power mode update keyword is determined to obtain the mode update correlation feature value; based on the difference variable between the computing power scheduling correlation feature value and the mode update correlation feature value, the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector is determined from the associated session interaction knowledge vector corresponding to the computing power scheduling business keyword and the associated session interaction knowledge vector corresponding to the cloud computing power mode update keyword. The computing resource business keywords corresponding to the target associated session interaction knowledge vector are used as the session interaction business topic keywords of the corresponding target session interaction knowledge information set. The computing resource session response result corresponding to the computing resource session data to be processed is determined based on the session interaction business topic keywords of the target session interaction knowledge information set.

2. The method as described in claim 1, characterized in that, The method further includes: Load the knowledge information sets of each associated session into the initial AI knowledge embedding network; The knowledge vector encoding module in the initial AI knowledge embedding network encodes the knowledge vectors of each associated session interaction knowledge information set to obtain the knowledge encoding of each associated session interaction. The knowledge cross-collision module in the initial AI knowledge embedding network cross-collides the encodings of the various related conversation interaction knowledge to obtain the related conversation interaction knowledge vector corresponding to the set computing power resource business keyword. The debugging method for the initial AI knowledge embedding network includes: Obtain each computing resource session data sample and the corresponding prior session response training annotation, and load the each computing resource session data sample into the original deep learning network; The knowledge vectors of each computing resource session data sample are encoded by the initial AI knowledge embedding network in the original deep learning network to obtain each computing resource session knowledge vector sample. The initial knowledge cross-collision module in the original deep learning network is used to cross-collision the knowledge vector samples of each computing resource session to obtain the cross-collision sample vector. The initial session response output module in the original deep learning network outputs a session response to the cross-collision sample vector to obtain the session response prediction result. The original deep learning network is optimized based on the prior session response training annotations and the session response prediction results to obtain an intermediate deep learning network; The intermediate deep learning network is used as the original deep learning network. The process of obtaining each computing resource session data sample and the corresponding prior session response training annotation, and loading each computing resource session data sample into the original deep learning network is repeated until the debugging process is completed and the target deep learning network is obtained. The initial AI knowledge embedding network is obtained based on the target knowledge vector encoding module and the target knowledge cross-collision module in the target deep learning network.

3. The method as described in claim 1, characterized in that, The process of identifying the target session interaction knowledge in the session data of the computing power resources to be processed, and obtaining a target session interaction knowledge information set, includes: Mine the session interaction semantic vector of the computing power resource session data to be processed, and perform feature preprocessing based on the session interaction semantic vector to obtain the session interaction semantic preprocessing vector; The preprocessed semantic vector of the conversation interaction is decoded for conversation requirements to obtain a conversation requirement decoding vector. Information is collected based on the conversation requirement decoding vector to obtain the target conversation interaction knowledge information set.

4. The method as described in claim 1, characterized in that, The step of determining the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector from the associated session interaction knowledge vectors corresponding to each computing resource business keyword based on the correlation feature values ​​of the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing resource business keyword includes: Determine the feature commonality score of the current session interaction knowledge vector and the associated session interaction knowledge vectors corresponding to each computing power resource business keyword, and obtain each relevance feature value; The target relevance feature value is determined from the various relevance feature values, and the associated session interaction knowledge vector corresponding to the target relevance feature value is used as the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector.

5. The method as described in claim 1, characterized in that, The step of determining the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector from the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the associated session interaction knowledge vector corresponding to the cloud computing power mode update keyword, based on the difference variable between the computing power scheduling related feature value and the mode update related feature value, includes: Determine the distinguishing variable between the computing power scheduling correlation feature value and the mode update correlation feature value. When the distinguishing variable is greater than a set variable value, perform a comparative analysis on the computing power scheduling correlation feature value and the mode update correlation feature value. When the computing power scheduling correlation feature value is greater than the mode update correlation feature value, the associated session interaction knowledge vector corresponding to the computing power scheduling business keyword is taken as the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector. When the correlation feature value of the mode update is greater than the correlation feature value of the computing power scheduling, the associated session interaction knowledge vector corresponding to the cloud computing power mode update keyword is taken as the target associated session interaction knowledge vector corresponding to the current session interaction knowledge vector. When the distinguishing variable is not greater than the set variable value, the general business topic keyword is used as the conversation interaction business topic keyword of the target conversation interaction knowledge information set.

6. The method as described in claim 1, characterized in that, The target session interaction knowledge information set includes at least two; determining the computing resource session response result corresponding to the computing resource session data to be processed based on the session interaction business topic keywords of the target session interaction knowledge information set includes: Determine the heat evaluation coefficients of the at least two target session interaction knowledge information sets with the computing power resource session data to be processed, and take the session interaction business topic keyword of the target session interaction knowledge information set with the largest heat evaluation coefficient as the computing power resource session response result corresponding to the computing power resource session data to be processed.

7. The method as described in claim 1, characterized in that, The method further includes: The target session interaction knowledge information set and the associated session interaction knowledge vectors corresponding to each computing power resource business keyword are loaded into a bidirectional long short-term memory network; The knowledge vector encoding module in the bidirectional long short-term memory network mines the current conversation interaction knowledge vector corresponding to the target conversation interaction knowledge information set. Based on the correlation feature values ​​between the current conversation interaction knowledge vector and the associated conversation interaction knowledge vectors corresponding to each computing power resource business keyword, the target associated conversation interaction knowledge vector corresponding to the current conversation interaction knowledge vector is determined from the associated conversation interaction knowledge vectors corresponding to each computing power resource business keyword. The computing power resource business keyword corresponding to the target associated conversation interaction knowledge vector is used as the conversation interaction business topic keyword of the corresponding target conversation interaction knowledge information set. The debugging method for the bidirectional long short-term memory network includes: Obtain a sample of conversational interaction knowledge information set and corresponding scenario conversation response training annotations; The associated conversational interaction knowledge vectors corresponding to the conversational interaction knowledge information set sample and the various computing power resource business keywords are loaded into the original bidirectional long short-term memory network for conversational keyword parsing to obtain the conversational interaction business prediction keywords corresponding to the conversational interaction knowledge information set sample. The original bidirectional long short-term memory network is optimized based on the predicted keywords of the conversational interaction service and the training annotations of the scenario conversational response to obtain an optimized bidirectional long short-term memory network. The optimized bidirectional long short-term memory network is used as the original bidirectional long short-term memory network. The process of obtaining the conversational interaction knowledge information set sample and the corresponding scenario conversation response training annotation is repeated until the debugging process is completed, and the bidirectional long short-term memory network is obtained.

8. A session response processing system, characterized in that, The method includes at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, causing the at least one processor to perform the method according to any one of claims 1-7.

Images