A Cross-Domain Book Recommendation Method and System Based on Neural Network

Abstract

The present invention provides a cross - domain book recommendation method and system based on a neural network, which relates to the technical field of intelligent book recommendation. It includes obtaining first information containing user reading history, retrieval behavior, research direction and paper information, and converting it into the initial interest topology structure and evolution trend of the user. Subsequently, a cross - domain interest transfer graph of the user is obtained through Markov clustering. Then, the interest transfer graph and the structured book text information are input into an extended probabilistic latent variable semantic model to obtain a semantic embedding expression integrating user interests. Through the correlation analysis between this expression and the interest transfer graph, the potential correlation score between the user and the book is calculated, and a cross - domain recommendation list is generated based on the ranking, realizing personalized and structured book recommendation. The present invention realizes the deep alignment of the interest transfer structure and the semantic space, significantly improving the accuracy and practicability of cross - domain recommendation.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent book recommendation, and more specifically, to a cross-domain book recommendation method and system based on a neural network. Background Art

[0002] In the current era of information explosion and rapid knowledge update, traditional book recommendation systems are mainly constructed based on classical methods such as collaborative filtering and content matching. Although certain effects have been achieved in the same-domain recommendation, when facing the cross-disciplinary and cross-domain information needs of users, these recommendation methods show obvious limitations. Especially among high-level user groups such as researchers and university teachers, the interest migration paths are highly heterogeneous and evolutionary, and traditional methods are difficult to capture complex interest migration trajectories and multi-dimensional semantic associations.

[0003] Therefore, there is an urgent need for a cross-domain book recommendation method and system based on a neural network to meet the comprehensive recommendation needs of users for diverse and highly relevant books. Summary of the Invention

[0004] The purpose of the present invention is to provide a cross-domain book recommendation method and system based on a neural network to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0005] In a first aspect, the present application provides a cross-domain book recommendation method based on a neural network, including:

[0006] Obtain first information, where the first information includes the user's reading history data, historical retrieval data, research direction labels, and information on papers published historically;

[0007] Based on the first information, convert it into an initial interest topology structure and the user's interest evolution trend, and perform Markov clustering processing based on the converted results to obtain a cross-domain interest migration map of the user;

[0008] Input the cross-domain interest migration map of the user and the preset structured information of book texts into an extended probabilistic latent variable semantic model for processing, and perform semantic embedding based on the initially obtained book theme distribution to obtain a semantic embedding expression of cross-domain books integrating the user's interests;

[0009] Perform correlation analysis on the semantic embedding expression and the cross-domain interest migration map of the user to obtain a cross-domain potential correlation scoring result between the user and the book;

[0010] Perform sorting processing based on the cross-domain potential correlation scoring result between the user and the book, and obtain a cross-domain book recommendation list based on the sorting result.

[0011] In a second aspect, the present application also provides a cross-domain book recommendation system based on a neural network, including:

[0012] An acquisition unit for acquiring first information, where the first information includes the user's reading history data, historical retrieval data, research direction tags, and information on papers published historically

[0013] A processing unit for converting the first information into an initial interest topology structure and the user's interest evolution trend, and performing Markov clustering processing based on the converted results to obtain the user's cross-domain interest migration map;

[0014] An embedding unit for inputting the user's cross-domain interest migration map and preset structured information of book texts into an extended probabilistic latent variable semantic model for processing, and performing semantic embedding based on the initially obtained book topic distribution to obtain a semantic embedding expression of cross-domain books integrating the user's interests;

[0015] An analysis unit for performing correlation analysis on the semantic embedding expression and the user's cross-domain interest migration map to obtain a cross-domain potential correlation scoring result between the user and the books;

[0016] A sorting unit for performing sorting processing based on the cross-domain potential correlation scoring result between the user and the books, and obtaining a cross-domain book recommendation list based on the sorting result.

[0017] The beneficial effects of the present invention are as follows:

[0018] By constructing an initial topology structure of the user's reading behavior and research interests, and combining Markov clustering to generate a cross-domain interest migration map, further integrating structured book semantic information, and obtaining multi-granularity semantic expressions through an extended probabilistic latent variable model and Poisson embedding; then, through a gated recurrent network, a heterogeneous attention autoencoder, and an exponential quantum graph network, deep association modeling between the user and the books is realized to obtain a potential scoring result; finally, through a combination of Bayesian sorting, restricted polynomial regression, and Lévy flight mechanism for sorting optimization, a recommendation result with both personalization and diversity is generated. The present invention realizes a deep alignment between the interest migration structure and the semantic space, significantly improving the accuracy and practicality of cross-domain recommendation.

[0019] Other features and advantages of the present invention will be described in the subsequent description, and, in part, will become apparent from the description, or will be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the structures specifically pointed out in the written description, claims, and drawings. Description of the Drawings

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0021] Figure 1 Schematic flowchart of the cross - domain book recommendation method based on neural network described in the embodiments of the present invention;

[0022] Figure 2 Schematic structural diagram of the cross - domain book recommendation system based on neural network described in the embodiments of the present invention.

[0023] In the figure: 701, acquisition unit; 702, processing unit; 703, embedding unit; 704, analysis unit; 705, sorting unit. Specific implementation manners

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0025] It should be noted that: similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, terms such as "first", "second", etc. are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0026] Embodiment 1

[0027] This embodiment provides a cross - domain book recommendation method based on neural network.

[0028] See Figure 1 , which shows that this method includes steps S1, S2, S3, S4, and S5.

[0029] Step S1: Obtain the first piece of information, which includes the user's reading history data, historical retrieval data, research direction tags, and information on historically published papers.

[0030] It can be understood that in this step, through the integration method of multi-source heterogeneous data, key feature information of the user's academic and reading behaviors is comprehensively collected to construct the basic dataset required for the user portrait. Specifically, the reading history data reflects the user's past literature reading preferences on various academic platforms, e-book platforms, or knowledge communities; the historical retrieval data reflects the user's interest orientation in actively obtaining information, such as keyword search records, query frequencies, and time period distributions; the research direction tags usually consist of research field tags actively declared by the user in the scientific research community or academic platform, and can also be indirectly extracted through paper keywords, conference classifications, etc.; and the information on historically published papers covers the user's past scientific research achievements, including structured information such as titles, abstracts, keywords, and the subject fields to which they belong. This step can break the recommendation limitations brought by a single interest tag, map the user's academic interest behavior into a dynamic representation basis with multi-dimensional features, and provide a highly accurate starting basis for subsequent interest modeling and migration. Especially in recommendation tasks involving user research direction migration or knowledge cross-border integration, this way of obtaining dynamic interests driven by behavior from static tags is more in line with the real user's interest evolution path, significantly improving the recommendation system's ability to grasp the user's intentions.

[0031] Step S2: Transform the first piece of information into an initial interest topology structure and the user's interest evolution trend, and perform Markov clustering processing based on the transformed results to obtain the user's cross-domain interest migration map.

[0032] It can be understood that this step not only performs multi-scale clustering on static interests but also accurately captures the user's behavior trend of spreading from the main research field to the marginal field through the dynamic modeling of the graph structure. Compared with traditional vector space models or classification methods, this graph model combined with Markov clustering provides stronger structural expression ability and migration pattern mining ability, improves the structural integrity and evolution interpretability of interest modeling, effectively enhances the system's perception and prediction ability of the user's long-term and cross-domain interest migration, and lays a foundation for the improvement of the recommendation algorithm in terms of time sensitivity and cross-border diversity. In this step, Step S2 includes Step S21, Step S22, Step S23, and Step S24.

[0033] Step S21: Perform self-organizing mapping network processing on the first piece of information. Specifically, through constructing a two-dimensional topology-preserving structure, non-linear dimensionality reduction and interest cluster mapping are performed on the high-dimensional interest vector containing the embedded representation of the first piece of information to obtain the user's initial interest topology structure.

[0034] It can be understood that this step first converts the user's reading history, search behavior, research direction, and published papers into a unified high-dimensional interest embedding vector through semantic encoding (BERT model). This vector contains both the content similarity at the semantic level and the contextual relationship between the temporal characteristics of user behavior and subject labels, and has rich information expression capabilities.

[0035] Secondly, this step introduces a self-organizing map network for unsupervised learning. The self-organizing map network constructs a neural network structure with two-dimensional grid nodes, in which each neuron represents the prototype of an interest cluster. Through competitive learning and neighborhood function update mechanism, the system finds the mapping position of each interest vector in the high-dimensional interest space, and maps vectors that are similar in semantics or behavior to adjacent or neighboring neurons in the two-dimensional topology, thereby achieving nonlinear dimensionality reduction while retaining the structural characteristics of the original interest distribution. This process allows different interest points to present clear clustering characteristics in the two-dimensional space, while maintaining the correlation and evolution direction between the original interests. The two-dimensional interest topological structure constructed in this step is a visual mapping result of the user's interest state. Each topological node represents a cluster interest center, and the adjacency relationship between nodes reflects the potential semantic transfer or behavior migration path. This structure provides a solid foundation for subsequent migration modeling and path identification based on graph methods.

[0036] Among them, the competition rules of the competitive learning mechanism are as follows:

[0037] in, is the weight of the node of interest closest to the input vector, is the learning rate, is the current user interest vector.

[0038] Among them, the update formula of the neighborhood function is as follows:

[0039] in, Representation Node In time step The weight of Representation Node In time step The weight of represents the learning rate, represents the weight calculated based on the neighborhood function, Represents the current user interest vector.

[0040] Step S22, performing discrete cosine transform processing on the initial interest topology structure of the user, wherein the long-term and short-term interest change trends are decoupled by extracting the frequency domain pattern of the time series weight signal in the same interest node trajectory, and the evolution trend of the user's interest among multiple preset research fields is obtained;

[0041] It can be understood that this step takes each interest node in the user's initial interest topology as a processing unit, and extracts its corresponding time series weight signal, which is usually composed of the user's behavior frequency or interest intensity within a time window, such as the number of views, search clicks, and literature citation frequency of a certain research direction. These behaviors are quantified into time series vectors. Then, these sequences are transformed into the frequency domain using discrete cosine transform, and its mathematical expression is that the time series performs the following transformation:

[0042] in, Indicates Discrete cosine transform coefficients, represents the total length of the input signal, The original input signal The time domain signal of sampling points, represents the time index of the input signal, represents the frequency index after transformation, It is pi.

[0043] The core function of this process is to map the user's behavior trajectory in a certain interest direction from the time domain to the frequency domain, and then identify the interest change patterns represented by different frequency bands: low-frequency coefficients reflect long-term stable interest tendencies, while high-frequency coefficients reveal the user's recent interest mutations or short-term behavioral interference. By setting the frequency domain energy threshold (for example, selecting the first few coefficients with cumulative energy reaching 90%), the system can effectively compress the input dimension and remove high-frequency noise, retaining only representative interest change features.

[0044] Furthermore, this step embeds these frequency domain features into the multidimensional research field space, and constructs an evolution tensor based on the frequency feature similarity between nodes to capture the user's interest shift path between multiple preset research fields. Unlike traditional time sliding window averaging or first-order difference methods, DCT has stronger expressiveness and compression capabilities, and can more clearly distinguish between trend interests (such as gradual growth) and cyclical preferences (such as staged regression), and has significant advantages when facing cyclical alternation or implicit migration signals.

[0045] This step realizes the frequency-domain modeling and decomposition of the user's interest time series behavior, enabling the system to structurally decouple the interest evolution process and accurately divide the composition of long-term stable interests and short-term fluctuating interests. This not only improves the forward-looking nature of cross-domain recommendations but also provides physically interpretable frequency tags for the subsequent construction of time weights in the interest migration graph, thereby achieving more refined and interpretable dynamic interest analysis.

[0046] Step S23: Perform domain label matching and nested community mining on the evolution trend of the user's interests among multiple preset research fields. Specifically, through an interest label attribution mechanism based on bidirectional label propagation, map each interest node to a preset research field to obtain an annotated interest evolution graph.

[0047] It can be understood that this step first introduces a domain label matching mechanism. This mechanism, based on the bidirectional label propagation method, aligns the user interest nodes with the preset research field labels. The specific operation is as follows: construct a two-layer graph structure, where one layer is the nodes in the user interest evolution graph, and the other layer is the set of preset research field labels (specific source: Web of Science subject classification). Then, through iterative propagation using the label propagation algorithm, on the one hand, propagate the known domain labels to the interest nodes, and on the other hand, reverse-propagate the labels of the high-weight nodes in the interest trajectory to the research field label layer, thereby realizing bidirectional attribution judgment. This mechanism not only utilizes the topological similarity of the graph structure but also combines the dynamic nature of node weights in the time evolution process.

[0048] After completing the interest label mapping, use a nested community mining algorithm to further identify the internal structural relationships in the interest graph. This process first divides the user interest graph into hierarchically nested sub-communities, each community corresponding to a research field or cross-field combination, and presenting an interest migration pattern where the user delves deeper or makes cross-layer jumps in these fields. Community division not only considers the evolutionary similarity and time consistency between nodes but also incorporates the frequency-domain features extracted in the previous steps, making the nested structure more consistent in terms of time and semantics.

[0049] The unique effect of this step is to transform the originally continuous interest evolution data into a graph representation with clear semantics and structure, enabling the subsequent recommendation model to perform multi-scale learning based on the "interest - domain" structure. At the same time, bidirectional label propagation avoids the problem of inaccurate attribution caused by sparse labels or domain intersections in traditional unidirectional matching, and the introduction of the nested community structure effectively improves the hierarchy and interpretability of interest structure modeling. The technical effect of this step is reflected in achieving an accurate alignment between the user interest evolution path and the domain knowledge structure, providing a semantic basis and structural support for accurate matching and reasoning in subsequent cross-domain recommendations.

[0050] Step S24: Perform Markov clustering on the annotated interest evolution graph. Specifically, by constructing a state transition matrix between user interest nodes and conducting random walk simulation, identify and summarize potential cross-domain interest migration paths to obtain the user's cross-domain interest migration graph.

[0051] It can be understood that in this step, first, a state transition matrix is constructed based on the state transition relationship between user interest nodes. In the graph, each node represents a specific interest point (such as a certain research field or a specific topic), and the edges between nodes represent the potential transition probability from one interest node to another. At this time, the elements of the state transition matrix reflect the probability that the user transfers from one interest state (research field, topic, etc.) to another. This transition probability not only considers the direct connection relationship between nodes but also adjusts the weights according to the user's historical behaviors (such as clicks, readings, retrievals, etc.), making the probability distribution more in line with the actual interest migration trajectory.

[0052] Next, conduct random walk simulation. Through the random walk algorithm, simulate the process of the user starting from a certain interest node and randomly walking to other nodes, and repeat this process multiple times. During the random walk process, based on the state transition matrix, each step is taken according to the transition probability between nodes. The purpose of the random walk simulation is to discover and identify potential cross-domain interest migration paths through multiple simulations. These paths reflect the migration rules of the user's interest from one field to another. By simulating different interest paths, it is possible to identify which fields have more frequent migrations and which fields have slower or less obvious interest transfers for users.

[0053] After completing the random walk simulation, based on the clustering algorithm in the Markov process (in this step, the K-means clustering is used), divide the interest nodes in the interest graph into several groups. These groups can effectively reveal the user's cross-domain interest migration graph, that is, the potential rules and preferences of the user's migration between different research fields. For example, some users may concentrate on a certain discipline field within a certain period and then turn to related or adjacent fields, while other users may show more scattered interest migration paths.

[0054] Through the Markov clustering method in this step, it is possible to reveal the user's potential interest migration patterns from a large-scale interest graph and provide a more refined interest trajectory analysis. Different from traditional keyword-based recommendation systems, Markov clustering introduces a dynamic transition process into the model, making the identification of cross-domain interest migration more in line with the actual user behavior pattern. At the same time, the random walk process helps the system overcome the limitations of a single path through a large number of random simulations, and can better explore the possible interest transfer directions of users, providing strong data support for subsequent personalized recommendations.

[0055] Step S3: Input the user's cross - domain interest transfer graph and the preset structured information of book texts into the extended probabilistic latent variable semantic model for processing, and perform semantic embedding based on the initially obtained book topic distribution to obtain the semantic embedding expression of cross - domain books integrating user interests;

[0056] It can be understood that this step combines two seemingly independent information sources, namely the cross - domain interest transfer graph and book topics, and uses the extended probabilistic latent variable semantic model to model the relationship between user interests and book texts. This fusion method not only improves the accuracy of interests but also can handle cross - domain interest transfer, making up for the limitation that traditional recommendation systems can only make recommendations within a single domain. In this step, step S3 includes step S31, step S32, and step S33.

[0057] Step S31: Perform extended probabilistic latent variable semantic model processing on the user's cross - domain interest transfer graph and the preset structured book information. Specifically, jointly model the book content, chapter structure, and domain labels through a preset hierarchical Bayesian inference mechanism to obtain the multi - granularity topic distribution of each book;

[0058] It can be understood that this step first constructs a joint input set integrating semantic features and interest evolution features based on the user's cross - domain interest transfer graph, combining the transfer paths and evolution trajectories between interest nodes. In this set, each interest node not only represents a research field or topic but also carries weights given by the user's historical behavior, such as reading frequency, duration, or interaction density under a specific topic. These features are quantitatively expressed through an interest state vector to drive subsequent topic modeling.

[0059] Next, input the above - mentioned set and the preset structured book information into the extended probabilistic latent variable semantic model. In this model, perform multi - layer semantic modeling processing by introducing a hierarchical Bayesian inference mechanism. This mechanism constructs a joint probability distribution among book content, chapter structure, and domain labels, and introduces latent variables in each layer to represent hidden topics. Specifically, in this step, perform parameter inference through Gibbs sampling to model the possible topics contained in each book at different levels (such as the whole book, chapter, and paragraph). During the topic extraction process, maintain the coherence of semantic connections between chapters through a context window mechanism, and use label information as a prior to guide the model to optimize the topic attribution.

[0060] Subsequently, based on the model inference results, a multi-granularity topic distribution for each book is generated, which includes the dominant semantics at the book level, the local topics at the chapter level, and the fine-grained semantic clusters of nested sub-topics. These topic distributions not only reflect the semantic diversity of the books themselves but also enable semantic comparison with the research field preferences shown by users in the interest migration path, laying a solid foundation for subsequent semantic embedding and matching recommendations.

[0061] Step S32: Input the multi-granularity topic distribution into a preset Poisson embedding model for processing. The preset Poisson embedding model generates a low-rank sparse embedding vector by simulating the high-dimensional Poisson counting characteristics of each book in the semantic space, obtaining a sparse semantic representation that matches the interest evolution structure.

[0062] It can be understood that in this step, the multi-granularity topic distribution of each book is first used as input to construct a set of high-dimensional semantic counting vectors. These counting vectors are based on the word frequency occurrences of each topic at each granularity level and reflect the "semantic intensity" characteristics of the book in the entire semantic space. Due to the hierarchical structure of the multi-granularity topics, the high-dimensional vectors not only contain the distribution frequencies of the main topics but also include the counting information of local sub-topics and marginal topics, forming a sparse and structured information expression basis.

[0063] Next, the above high-dimensional semantic counting information is input into the preset Poisson embedding model for processing. The Poisson embedding model is a type of probabilistic graphical model with a Poisson distribution as the generation mechanism and is suitable for processing high-dimensional discrete counting data. In actual operation, it is assumed that the semantic counting of each book is a Poisson distribution sample generated by a latent low-dimensional embedding vector. Specifically, the model factorizes the original high-dimensional semantic counting matrix by introducing a low-rank semantic space matrix, decoupling the basic topic dimensions in the semantic space and the latent variable expression of the book semantic features. This embedding process uses maximum likelihood estimation to solve the model parameters to ensure that the generated embedding vectors have lower dimensions and stronger sparsity while maintaining the original semantic information. Among them,

[0064] Through this processing process, low-dimensional sparse embedding vectors corresponding to each book can be obtained. These vectors not only maintain consistency with the original topic distribution in the semantic space but also have good matching characteristics for the semantic migration relationship formed by the interest evolution path in terms of structure. The finally output sparse semantic representation not only compresses redundant semantic dimensions but also highlights the semantic core areas with potential migration value, enabling the subsequent recommendation model to more efficiently capture the deep matching relationship between user interests and book content.

[0065] Step S33: Perform adaptive correlation entropy embedding processing based on the sparse semantic expression and the user's jump path in the interest transfer graph. Specifically, construct a user-path weighted semantic covariance matrix and minimize its information entropy offset to obtain a personalized cross-domain semantic embedding expression aligned with the interest transfer pattern.

[0066] It can be understood that in this step, based on the sparse semantic expression vector obtained in the previous step and the jump path information of the user in the interest transfer graph, a user-path weighted semantic covariance matrix is constructed. This covariance matrix not only captures the statistical correlation between different dimensions in the semantic embedding space but also assigns personalized weights to the semantic relationships between each pair of books by integrating the jump trajectories in the interest transfer graph. Specifically, for book pairs involved in the user's jump from one research field to another, higher path importance weights are assigned according to indicators such as jump frequency, path length, and node similarity, making the covariance matrix more closely aligned with the actual interest evolution dynamics.

[0067] Subsequently, on the basis of this weighted semantic covariance structure, a correlation entropy embedding mechanism is introduced for optimization. The core idea of this method is to minimize the information entropy offset between the semantic covariance generated by the embedding representation guided by the user path and the expected distribution, ensuring that the embedding space is not only sparse but also can truly reflect the semantic connections driven by interest transfer. This optimization process uses the KL divergence entropy in information theory as the metric standard, and by iteratively minimizing the uncertainty in the embedding space, the semantic vectors involved in the user interest transfer become more concentrated and discriminative, thus constructing a semantic representation highly consistent with the individual interest evolution trajectory.

[0068] The final output is a set of personalized cross-domain semantic embedding expressions aligned with the user's interest transfer pattern. These embeddings not only retain the advantages of semantic sparsity and low-dimensional expression but also precisely adapt to the transfer path of the user in the interest map in terms of structure, forming a deep expression pattern of the synergy among the user, semantics, and domain.

[0069] Step S4: Perform correlation analysis on the semantic embedding expression and the user's cross-domain interest transfer graph to obtain the cross-domain potential association scoring results between the user and the books;

[0070] It can be understood that in this step, by fusing and analyzing semantic embeddings with the interest transfer graph structure, not only is an accurate characterization of the potential interest associations between users and books in different fields achieved, but also the latent demands that have not been explicitly expressed in the interest evolution path are effectively mined. Compared with traditional recommendation methods based on content or collaborative filtering, this method can capture the deep - level dynamic evolution of user interests at both the semantic and structural levels, greatly improving the accuracy and personalization of cross - domain recommendations. In this step, step S4 includes step S41, step S42, step S43, and step S44.

[0071] Step S41: Perform gated recurrent hybrid network processing on the semantic embedding expression and the user's cross - domain interest transfer graph. Among them, by introducing a gating mechanism to control the information flow of the semantic embedding sequence and the interest graph jump sequence, and capturing the interest diffusion characteristics at the long - short - term level, a hybrid interest evolution trajectory between the user and the book is obtained.

[0072] It can be understood that in this step, the sparse semantic embedding sequence personalized for the user is first used as the main input channel, and it is time - aligned with the node jump sequence (i.e., the evolution path of the user's interest among different fields) contained in the user interest transfer graph. By using timestamps or event sequence numbers, it is ensured that the two types of information have a consistent evolution order. Then, a two - channel recurrent network structure with a gating mechanism is introduced: the main channel uses a GRU (Gated Recurrent Unit) unit with an update gate and a reset gate to model the semantic embedding information, and the secondary channel uses another GRU structure to model the interest jump sequence. The two are fused through a cross - gating layer.

[0073] In the cross - gating layer, a trainable gating weight vector is used to dynamically weight - fuse the hidden states of the main channel and the secondary channel. Its weight distribution is adjusted according to the joint index of the semantic context change rate and the interest jump frequency at the current time step, so as to ensure that the model can dynamically switch between semantic - dominated and structure - dominated. This design can effectively avoid the problems of information loss or over - memorization in long - sequence modeling, and at the same time achieve fast response to information at the interest evolution mutation points (such as jumping from one research field to another).

[0074] To further enhance the model's ability to perceive interest diffusion characteristics at different time scales, the gated recurrent hybrid network introduces a hierarchical recurrent structure in its upper layer, that is, capturing the long - term interest evolution trend of the user within a long - time window and capturing local hot - spot interest fluctuations within a short - time window. This hierarchical capture mechanism generates the final hybrid interest evolution trajectory representation by parallel - computing the recurrent hidden states at two different time scales and aggregating them with attention weighting.

[0075] Step S42: Perform heterogeneous attention autoencoder processing based on the mixed interest evolution trajectory. Specifically, by constructing a multi-channel attention graph based on preset user behavior types, enhance the discriminability of interest expression in the dimension of behavior heterogeneity to obtain a structurally enhanced user-book interaction representation.

[0076] It can be understood that in this step, based on the mixed interest evolution trajectory obtained in the previous stage, the behavior nodes corresponding to the user at each time step are classified and mapped to construct a multi-channel heterogeneous attention graph. In this graph, each channel represents a type of user behavior. The edge weights between the user and the books under different channels reflect the importance of this behavior in this semantic context, and the nodes are represented by the semantic embedding vectors and interest jump sequences in the mixed trajectory.

[0077] Next, through the heterogeneous attention mechanism, introduce an independent attention scoring network for each type of behavior to calculate the importance coefficient of the user node for the adjacent book node in the current behavior channel. The calculation of the attention score not only considers the intensity of the user's historical behavior (such as frequency, duration), but also introduces multiple factors such as domain label consistency, semantic embedding similarity, and context weights of domain jumps to construct an attention weight tensor with behavior semantic perception ability. In the autoencoder structure, the attention weights of each channel are used to guide the aggregation process of adjacent node information in the encoding stage, so that the final embedding of each user node not only retains its original semantic features but also integrates its interaction preference features under different behavior channels. Among them, the process of calculating the importance coefficient of the user node for the adjacent book node in the current behavior channel is as follows: First, input the user interest sequence into two parallel RNN structures simultaneously:

[0078]

[0079]

[0080] where is the short-term interest hidden state at the th moment, represents the user interest input feature at the th moment, represents the short-term interest hidden state at the th moment, represents the long-term interest hidden state at the th moment, represents the long-term interest hidden state at the th moment, represents the recurrent neural network structure for modeling short-term interest dynamics, and represents the recurrent neural network structure for modeling long-term interest evolution.

[0081] Then, importance coefficient calculation is performed through weighted fusion. Among them, the calculation formula for the importance coefficient is:

[0082] Among them, represents the importance coefficient, represents the attention score related to the th item, represents the number of all items considered, represents the summation index variable, and represents the score of the th candidate element.

[0083] After encoding is completed, the user embeddings generated by each channel will be fed into the decoder for reconstruction. By minimizing the difference between the original graph structure (such as the connection strength of user-book edges) and the reconstructed graph, the model is driven to learn an embedding representation that is more sensitive to the user's true behavioral preferences. Finally, the reconstructed vectors of each channel are weighted and fused to form a structure-enhanced user-book interaction representation.

[0084] Step S43: Perform exponential quantum graph network processing based on the structure-enhanced user-book interaction representation. By constructing a quantum state extended path graph with non-commutative adjacency characteristics and introducing an exponential decay coefficient and a quantum state coupling mechanism during node propagation, a cross-domain semantic coupling score vector between the user and the book is obtained;

[0085] It can be understood that in this step, a "quantum state extended path graph" is first constructed based on the structure-enhanced user-book interaction representation. In this graph, user and book nodes are regarded as quantum systems in a certain semantic superposition state. The edges between nodes no longer only represent adjacency relationships, but an adjacency weight matrix with non-commutative characteristics is introduced, allowing semantic propagation to occur under an asymmetric structure, reflecting the non-uniform perception of the user towards different books.

[0086] In the node propagation mechanism, the quantum graph network introduces two core mechanisms: one is the exponential decay coefficient, which is used to simulate the information decay phenomenon that occurs as the path depth increases during semantic propagation in the graph. That is, the longer the propagation path, the smaller the coupling effect between nodes. The form of this coefficient is where represents the path length, is the attenuation intensity factor, which is used to control the suppression intensity of long-term dependencies; the second is the quantum state coupling mechanism, which is used to capture the coupling degree between the user interest state and the book semantic state in the superposition space. It is used to control the suppression intensity of long-term dependencies; the second is the quantum state coupling mechanism, which is used to capture the coupling degree between the user interest state and the book semantic state in the superposition space. This mechanism constructs a coupling tensor based on the Heisenberg quantum coupling model to quantify the intensity and directionality of the mutual influence between nodes, allowing interference terms and superposition interference to occur between multiple interest topics, and is particularly suitable for expressing the fuzzy interest preferences of users among interdisciplinary books.

[0087] In the iterative propagation process of the graph network, each round of information update is affected not only by the features of adjacent nodes, but also by the attenuation correction of the path weight and the interaction of quantum states. Finally, by aggregating the quantum state interaction information between all nodes, a cross-domain semantic coupling score vector is generated. This score vector can finely reflect the implicit interest coupling intensity of users for books in different fields and different topics, thus providing a more discriminative feature basis for downstream recommendation ranking.

[0088] Step S44: Perform potential score result estimation processing according to the cross-domain semantic coupling score vector. Specifically, a collaborative inference mechanism with a migration normalization term is constructed to perform soft filling and boundary convergence optimization on the score matrix, and a cross-domain potential association score result between the user and the book is obtained.

[0089] It can be understood that in this step, first, a user-book score matrix is constructed, where the missing items indicate that the user has not interacted with the book yet. The system uses the coupling score vector obtained in the previous step as the initial soft score input to guide the collaborative filtering framework for filling. To improve the migration robustness, the collaborative inference mechanism introduces a migration normalization term, which normalizes the score value according to the jump intensity of the user in the interest migration graph to prevent a certain interest source from overly dominating the score generation process. The form of the normalization function is as follows:

[0090] where, is the normalized potential score, is the semantic coupling score, is the user's jump intensity from node to other nodes in the graph.

[0091] Subsequently, the system uses collaborative matrix soft filling methods (such as matrix completion based on kernel regression, low-rank filling with graph regularization, etc.) to optimize and solve the normalized rating matrix. During the solution process, a boundary convergence optimization mechanism is introduced. By setting a dynamically adjusted convergence threshold and local loss constraints, it is ensured that the rating prediction does not fall into overfitting or propagation distortion. Especially when dealing with cold-start users or less-interacted books, the boundary mechanism can stabilize the rating output by fusing the semantic coupling features of neighboring users or books.

[0092] Finally, the cross-domain latent association rating results output by this step not only fully retain the preference patterns of users on the interest evolution path but also take into account the global consistency of the rating matrix structure, providing an accurate and interpretable numerical basis for subsequent book ranking recommendations.

[0093] Step S5: Perform sorting processing based on the cross-domain latent association rating results between users and books, and obtain a cross-domain book recommendation list based on the sorting results.

[0094] It can be understood that in this step, through a multi-factor sorting and dynamic weighting mechanism, while ensuring the accuracy of semantic coupling, the personalized expression ability and domain expansion ability of the recommended content are improved, making the recommendation system have stronger interpretability, adaptability, and user knowledge gain potential. In this step, step S5 includes step S51, step S52, and step S53.

[0095] Step S51: Perform Bayesian rank sorting processing according to the latent rating results. By constructing a hierarchical sorting structure model of rating confidence, estimate the preference levels of each book on the basis of maximizing the posterior probability of the rating distribution, and obtain a preliminary sorted list of candidate books.

[0096] It can be understood that in this step, first, the rating result of each user-book pair is regarded as an observed sample generated from a hidden preference level, and this preference level itself, as a latent variable to be estimated, belongs to a discrete level space (such as "very interested", "general", "not interested", etc.). The system further introduces a confidence factor for each rating value, which is comprehensively estimated by indicators such as the semantic coupling degree of the rating, the number of interest jump paths, and semantic sparsity, serving as the prior uncertainty modeling basis for the rating accuracy.

[0097] Among them, the confidence factor is:

[0098] Among them, represents the confidence factor of the rating, represents the similarity between the user's interest and the book content, represents the frequency of the user's interest migration, Indicates the sparsity of the match between the interest and the book theme.

[0099] In the model construction stage, the system uses a hierarchical Bayesian modeling strategy to model the rating generation process as a three-layer structure: the first layer models the prior distribution of user preferences (a Beta distribution is adopted in this step); the second layer models the variance structure of the rating confidence, and the third layer models the observed values of the actual ratings, regarded as sampling results generated by noise perturbation based on the preference level. On this basis, the system uses the variational inference method to maximize the posterior probability estimation of the level of each candidate book, so as to infer its ranking level under the current user interest preference.

[0100] Different from the traditional point-value ranking, Bayesian ranking fully considers the confidence difference and potential uncertainty of ratings during the estimation process, and has stronger fault tolerance and ranking stability. It is especially suitable for dealing with sparse ratings or semantic drift in cross-domain recommendation systems to ensure that the preliminary sorted list is more in line with the actual preferences of users.

[0101] Step S52: Perform restricted least polynomial regression processing according to the preliminary sorted list of the candidate books. Specifically, by establishing a low-order non-linear mapping model between the sorting sequence and the semantic similarity function, and adding a structural smoothing penalty term to correct extreme sorting offsets, a sorted trend result after semantic calibration is obtained;

[0102] It can be understood that in this step, the sorting sequence (i.e., the preliminary sorted list of candidate books) is used as the independent variable, and the semantic similarity function (the similarity between the book content and the user interest calculated by multi-dimensional semantic embedding, etc.) is used as the dependent variable to construct a low-order polynomial regression model. The purpose of this regression model is to capture the non-linear mapping law between them by fitting the relationship between the preliminary sorted list and the semantic similarity. For example, some books may be ranked ahead in the preliminary sorted list due to high ratings, but their semantic similarity is low. The regression model will adjust their sorting positions according to this deviation to ensure that the final sorting is more consistent with the user's interest and semantic matching degree.

[0103] Among them, the low-order polynomial regression model is as follows:

[0104] Among them, represents the dependent variable, represents the independent variable, represents the regression coefficient, represents the error term.

[0105] Secondly, in the regression process, this step introduces a structural smoothing penalty term to correct extreme deviations in the ranking. The role of this penalty term is to limit the complexity of the regression model, avoid overfitting to noise or outliers in the data, and ensure a smooth transition of the ranking adjustment process within an effective range. For example, when the preliminary ranking of some books has a very low degree of match with semantic similarity, the structural smoothing penalty term will moderately control their ranking changes, prevent extreme position deviations in the final ranking, and thus avoid the imbalance of the overall ranking caused by individual outliers.

[0106] Step S53: Process the ranking trend result after semantic calibration using a Lévy flight sorter. Specifically, the Lévy flight sorter simulates the non-Gaussian jump distribution characteristics of the user's cognitive path and randomly inserts book nodes with a relatively large feature space distance to obtain a cross-domain book recommendation list with enhanced diversity.

[0107] It can be understood that the Lévy flight in this step is a random process. Its characteristic is that the jump length follows a Lévy distribution, which has heavy-tailed characteristics, meaning it can simulate long-distance jumps or non-Gaussian jumps. Compared with the traditional Gaussian random process, the Lévy flight can generate longer jumps. This enables it to better represent the long-distance jumps in the user's interest exploration behavior when simulating the user's cognitive path, that is, the user may suddenly become interested in some books with a relatively large feature space distance. Such behavior often does not conform to the conventional nearest neighbor recommendation mode but reflects the exploratory characteristics of human cognition. For example, after being exposed to knowledge in multiple related fields, the user's interests may span multiple different fields or topics, resulting in a relatively broad preference.

[0108] In specific operations, the Lévy flight sorter first processes the ranking trend result after semantic calibration. Based on the characteristics of the Lévy flight, it simulates the process of the user jumping from a known interest field to a more distant interest node and adjusts the order of the books in the recommendation list. In this way, the system will randomly insert book nodes with a relatively large feature space distance. These books may not be the most relevant to the current user's interest field, but by simulating long-distance jumps in the cognitive process, they can bring new interest inspiration and knowledge expansion. Such insertions can increase the diversity of the recommendation list, enabling users to not only be exposed to closely related books but also discover cross-domain knowledge and information, thus enriching their reading experience.

[0109] Embodiment 2:

[0110] As Figure 2 shown, this embodiment provides a cross-domain book recommendation system based on a neural network. Refer to Figure 2The system includes an acquisition unit 701, a processing unit 702, an embedding unit 703, an analysis unit 704, and a sorting unit 705.

[0111] The acquisition unit 701 is configured to acquire first information, where the first information includes the user's reading history data, historical retrieval data, research direction tags, and information on historically published papers

[0112] The processing unit 702 is configured to convert the first information into an initial interest topology structure and the user's interest evolution trend, and perform Markov clustering processing based on the converted results to obtain a cross-domain interest migration map of the user;

[0113] The embedding unit 703 is configured to input the cross-domain interest migration map of the user and the preset structured information of the book text into an extended probabilistic latent variable semantic model for processing, and perform semantic embedding based on the initially obtained book topic distribution to obtain a semantic embedding expression of the cross-domain book integrating the user's interests;

[0114] The analysis unit 704 is configured to perform correlation analysis on the semantic embedding expression and the cross-domain interest migration map of the user to obtain a cross-domain potential association scoring result between the user and the book;

[0115] The sorting unit 705 is configured to perform sorting processing based on the cross-domain potential association scoring result between the user and the book, and obtain a cross-domain book recommendation list based on the sorting result.

[0116] It should be noted that regarding the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0117] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0118] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A cross - domain book recommendation method based on neural network, characterized in that Including: Obtain first information, where the first information includes the user's reading history data, historical retrieval data, research direction tags, and information on historically published papers; Based on the first information, transform it into an initial interest topological structure and the user's interest evolution trend, and perform Markov clustering processing based on the transformed results to obtain the user's cross-domain interest migration map; Input the user's cross-domain interest migration map and the preset structured information of the book text into an extended probabilistic latent variable semantic model for processing, and perform semantic embedding based on the initially obtained book topic distribution to obtain the semantic embedding expression of the cross-domain book integrating the user's interests. Here, the extended probabilistic latent variable semantic model is a model that realizes the topic distribution through joint modeling of three layers of latent variables of books, chapters, and paragraphs, realizes parameter inference through Gibbs sampling, uses label information as a prior to optimize topic attribution, and constrains the topic differences between adjacent paragraphs through a context window mechanism to generate the multi-granularity topic distribution of each book; Perform correlation analysis on the semantic embedding expression and the user's cross-domain interest migration map to obtain the cross-domain potential association scoring result between the user and the book; Perform sorting processing based on the cross-domain potential association scoring result between the user and the book, and obtain a cross-domain book recommendation list based on the sorting result; Among them, based on the first information, transform it into an initial interest topological structure and the user's interest evolution trend, and perform Markov clustering processing based on the transformed results, including: Perform self-organizing mapping network processing on the first information. Among them, perform non-linear dimensionality reduction and interest cluster mapping on the high-dimensional interest vector containing the embedded representation of the first information by constructing a two-dimensional topology-preserving structure to obtain the user's initial interest topological structure; Perform discrete cosine transform processing on the user's initial interest topological structure. Among them, decouple the long-term and short-term interest change trends by extracting the frequency domain pattern of the time series weight signal in the same interest node trajectory to obtain the evolution trend of the user's interests among multiple preset research fields; Perform domain label matching and nested community mining processing on the evolution trend of the user's interests among multiple preset research fields. Among them, map each interest node to a preset research field through an interest label attribution mechanism based on bidirectional label propagation to obtain the annotated interest evolution map; Perform Markov clustering processing on the annotated interest evolution map. Among them, identify and summarize the potential cross-domain interest migration paths by constructing a state transition matrix between user interest nodes and performing random walk simulation to obtain the user's cross-domain interest migration map.

2. The cross-domain book recommendation method based on a neural network according to claim 1, wherein Input the user's cross-domain interest migration map and the preset structured information of the book text into an extended probabilistic latent variable semantic model for processing, and perform semantic embedding based on the initially obtained book topic distribution, including: Input the user's cross-domain interest migration map and the preset structured book information into an extended probabilistic latent variable semantic model for processing. Among them, jointly model the book content, chapter structure, and domain labels through a preset hierarchical Bayesian inference mechanism to obtain the multi-granularity topic distribution of each book; Input the multi-granularity topic distribution into a preset Poisson embedding model for processing. The preset Poisson embedding model generates a low-rank sparse embedding vector by simulating the high-dimensional Poisson counting characteristics of each book in the semantic space, and obtains a sparse semantic expression matching the interest evolution structure. Perform adaptive correlation entropy embedding processing based on the sparse semantic expression and the jump path of the user in the interest transfer graph. By constructing a semantic covariance matrix weighted by the user path and minimizing its information entropy offset, a personalized cross-domain semantic embedding expression aligned with the interest transfer pattern is obtained.

3. The cross-domain book recommendation method based on a neural network according to claim 1, wherein, Conduct correlation analysis on the semantic embedding expression and the user's cross-domain interest transfer graph, including: Perform gated recurrent hybrid network processing based on the semantic embedding expression and the user's cross-domain interest transfer graph. By introducing a gating mechanism to control the information flow between the semantic embedding sequence and the interest graph jump sequence, and capturing the interest diffusion characteristics at the long-term and short-term levels, a hybrid interest evolution trajectory between the user and the book is obtained. Perform heterogeneous attention autoencoder processing based on the hybrid interest evolution trajectory. By constructing a multi-channel attention map based on the preset user behavior types, the discriminability of the interest expression is enhanced in the dimension of behavior heterogeneity, and a structure-enhanced user-book interaction representation is obtained. Perform exponential quantum graph network processing based on the structure-enhanced user-book interaction representation. By constructing a quantum state extended path graph with non-commutative adjacency characteristics, an exponential decay coefficient and a quantum state coupling mechanism are introduced during the node propagation process, and a cross-domain semantic coupling score vector between the user and the book is obtained. Perform latent score result estimation processing based on the cross-domain semantic coupling score vector. By constructing a collaborative inference mechanism with a migration normalization term to perform soft filling and boundary convergence optimization on the score matrix, a cross-domain latent association score result between the user and the book is obtained.

4. The cross-domain book recommendation method based on a neural network according to claim 1, characterized in that Perform sorting processing based on the cross-domain latent association score result between the user and the book, and obtain a cross-domain book recommendation list based on the sorting result, including: Perform Bayesian rank sorting processing based on the latent association score result. By constructing a hierarchical sorting structure model of the score confidence, the preference level of each book is estimated based on maximizing the posterior probability of the score distribution, and a preliminary sorted list of candidate books is obtained. Perform restricted least polynomial regression processing based on the preliminary sorted list of candidate books. By establishing a low-order non-linear mapping model between the sorting sequence and the semantic similarity function, and adding a structure smoothing penalty term to correct extreme sorting offsets, a sorted trend result with semantic calibration is obtained. Perform Lévy flight sorter processing based on the sorted trend result with semantic calibration. By simulating the non-Gaussian jump distribution characteristics of the user's cognitive path with a Lévy flight sorter and randomly inserting book nodes with a large feature space distance, a cross-domain book recommendation list with enhanced diversity is obtained.

5. A cross-domain book recommendation system based on a neural network, characterized in that, Including: An acquisition unit for acquiring first information, where the first information includes the user's reading history data, historical retrieval data, research direction labels, and information on papers published historically. A processing unit, configured to transform the first information into an initial interest topology structure and the user's interest evolution trend, and perform Markov clustering processing based on the transformed results to obtain a cross-domain interest migration map of the user; An embedding unit, configured to input the cross-domain interest migration map of the user and the preset structured information of the book text into an extended probabilistic latent variable semantic model for processing, and perform semantic embedding based on the initially obtained book topic distribution to obtain a semantic embedding expression of the cross-domain book integrating the user's interests, where the extended probabilistic latent variable semantic model is a model that realizes the topic distribution through the joint modeling of three layers of latent variables of books, chapters, and paragraphs, realizes parameter inference through Gibbs sampling, uses label information as a prior to optimize the topic attribution, and constrains the topic difference between adjacent paragraphs through a context window mechanism to generate a multi-granularity topic distribution of each book; An analysis unit, configured to perform correlation analysis on the semantic embedding expression and the cross-domain interest migration map of the user to obtain a cross-domain potential association scoring result between the user and the book; A sorting unit, configured to perform sorting processing based on the cross-domain potential association scoring result between the user and the book, and obtain a cross-domain book recommendation list based on the sorting result; Wherein, the processing unit includes: A first processing subunit, configured to perform self-organizing mapping network processing on the first information, where non-linear dimensionality reduction and interest cluster mapping are performed on the high-dimensional interest vector containing the first information embedding representation by constructing a two-dimensional topology-preserving structure to obtain the user's initial interest topology structure; A second processing subunit, configured to perform discrete cosine transform processing on the user's initial interest topology structure, where the time-frequency pattern of the time-series weight signal is extracted in the same interest node trajectory to decouple the long-term and short-term interest change trends to obtain the evolution trend of the user's interests among multiple preset research fields; A third processing subunit, configured to perform domain label matching and nested community mining processing on the evolution trend of the user's interests among multiple preset research fields, where each interest node is mapped to a preset research field through an interest label attribution mechanism based on bidirectional label propagation to obtain an annotated interest evolution graph; A fourth processing subunit, configured to perform Markov clustering processing on the annotated interest evolution graph, where the state transition matrix between user interest nodes is constructed and random walk simulation is performed to identify and summarize the potential interest migration paths between fields to obtain the cross-domain interest migration map of the user.

6. The cross-domain book recommendation system based on a neural network according to claim 5, wherein The embedding unit includes: A first embedding subunit, configured to input the cross-domain interest migration map of the user and the preset structured book information into an extended probabilistic latent variable semantic model for processing, where the book content, chapter structure, and domain labels are jointly modeled through a preset hierarchical Bayesian inference mechanism to obtain a multi-granularity topic distribution of each book; A second embedding subunit, configured to input the multi-granularity topic distribution into a preset Poisson embedding model for processing, wherein the preset Poisson embedding model generates a low-rank sparse embedding vector by simulating the high-dimensional Poisson counting characteristics of each book in the semantic space, and obtains a sparse semantic representation matching the interest evolution structure; A third embedding subunit, configured to perform adaptive correlation entropy embedding processing according to the sparse semantic representation and the user's jump path in the interest transfer graph, wherein by constructing a user-path weighted semantic covariance matrix and minimizing its information entropy shift, a personalized cross-domain semantic embedding expression aligned with the interest transfer pattern is obtained.

7. The cross-domain book recommendation system based on a neural network according to claim 5, wherein The analysis unit includes: A first analysis subunit, configured to perform gated recurrent hybrid network processing according to the semantic embedding expression and the user's cross-domain interest transfer graph, wherein by introducing a gating mechanism to control the information flow between the semantic embedding sequence and the interest graph jump sequence, and capturing the interest diffusion characteristics at the long-term and short-term levels, a hybrid interest evolution trajectory between the user and the book is obtained; A second analysis subunit, configured to perform heterogeneous attention autoencoder processing according to the hybrid interest evolution trajectory, wherein by constructing a multi-channel attention map based on a preset user behavior type, the discriminability of the interest expression is enhanced in the dimension of behavior heterogeneity, and a structure-enhanced user-book interaction representation is obtained; A third analysis subunit, configured to perform exponential quantum graph network processing according to the structure-enhanced user-book interaction representation, by constructing an exponential decay coefficient and quantum state coupling mechanism in the node propagation process through a quantum state expansion path graph with non-commutative adjacency characteristics, a cross-domain semantic coupling score vector between the user and the book is obtained; A fourth analysis subunit, configured to perform potential score result estimation processing according to the cross-domain semantic coupling score vector, wherein by constructing a collaborative inference mechanism with a migration normalization term to perform soft filling and boundary convergence optimization on the score matrix, a cross-domain potential association score result between the user and the book is obtained.

8. The cross-domain book recommendation system based on a neural network according to claim 5, wherein The sorting unit includes: A first sorting subunit, configured to perform Bayesian rank sorting processing according to the potential association score result, by constructing a hierarchical sorting structure model of the score confidence, and estimating the preference level of each book based on maximizing the posterior probability of the score distribution, a preliminary sorted list of candidate books is obtained; A second sorting subunit, configured to perform restricted least polynomial regression processing according to the preliminary sorted list of candidate books, wherein by establishing a low-order non-linear mapping model between the sorting sequence and the semantic similarity function, and adding a structure smoothing penalty term to correct extreme sorting offsets, a sorted trend result after semantic calibration is obtained; A third sorting subunit, configured to perform Lévy flight sorter processing according to the sorted trend result after semantic calibration, wherein by simulating the non-Gaussian jump distribution characteristics of the user's cognitive path through the Lévy flight sorter, book nodes with a relatively large distance in the feature space are randomly inserted, and a cross-domain book recommendation list with enhanced diversity is obtained.

Images