CN112214687A - Paper recommendation method, system and medium for temporal perception academic information - Google Patents
Paper recommendation method, system and medium for temporal perception academic information Download PDFInfo
- Publication number
- CN112214687A CN112214687A CN202011050718.3A CN202011050718A CN112214687A CN 112214687 A CN112214687 A CN 112214687A CN 202011050718 A CN202011050718 A CN 202011050718A CN 112214687 A CN112214687 A CN 112214687A
- Authority
- CN
- China
- Prior art keywords
- user
- paper
- thesis
- data
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a paper recommendation method, a system and a medium for temporal perception academic information, wherein the method comprises the following steps: extracting the characteristics of the user information and the thesis information to obtain user characteristic data, the thesis characteristic data and grading data; training a neural network according to the user characteristic data, the thesis characteristic data and the grading data to obtain an initial recommendation model; carrying out dynamic weighting processing on the initial recommendation model through a temporal perception function to obtain a target recommendation model; and determining a recommended paper according to the target recommendation model. The embodiment of the invention improves the accuracy and the practicability of the recommendation system, and can be widely applied to the technical field of content recommendation.
Description
Technical Field
The invention relates to the technical field of content recommendation, in particular to a paper recommendation method, a system and a medium for temporal perception academic information.
Background
The temporal perception function is a time change rule function, and is mainly used for describing how time changes, including the time change rate, so as to obtain a change track, generally including linearity, logistic regression, exponential type and the like. Can be used to characterize a movie, an article, a short video, a change in popularity, a change in user interest, etc.
With the rapid development of social networks, the communication between scholars and scholars is increasing. However, in the era of information explosion, it is a significant challenge how to mine meaningful information from massive amounts of data. The recommendation system plays a very important role in relieving information overload, and is widely applied to various online services including social networks. Therefore, communication between scholars is more frequent, resulting in a large number of academic papers. The interest in finding people from a variety of papers is the key to modern academic paper recommendation systems.
At present, academic paper recommendation algorithms mainly include collaborative filtering algorithms and content-based recommendation methods. Wherein the collaborative filtering algorithm in turn comprises user-based and item-based collaborative filtering. According to the traditional collaborative filtering algorithm based on the neighborhood, the history of the similarity between users or articles is calculated according to the behavior data of the users, then the users or articles with high similarity are divided into the same group of neighborhoods, and finally, the neighbors recommend each other.
The algorithm can utilize feedback information of similar users to find potential interests of the users, but the problems of cold start, sparseness, short timeliness and the like of data also exist. In addition, time is a very important factor. It is known that if an academic paper was published decades ago, its effect will be weak, which means that it is of less reference value.
Therefore, a method for solving the problems of cold start, sparseness and the like of data and effectively controlling the timeliness of the data is needed.
Disclosure of Invention
In view of this, embodiments of the present invention provide a paper recommendation method, system and medium for temporal awareness academic information with high accuracy and high practicability.
The invention provides a paper recommendation method of temporal perception academic information, which comprises the following steps:
extracting the characteristics of the user information and the thesis information to obtain user characteristic data, the thesis characteristic data and grading data;
training a neural network according to the user characteristic data, the thesis characteristic data and the grading data to obtain an initial recommendation model;
carrying out dynamic weighting processing on the initial recommendation model through a temporal perception function to obtain a target recommendation model;
and determining a recommended paper according to the target recommendation model.
In some embodiments, the extracting features of the user information and the thesis information to obtain user feature data, thesis feature data, and score data includes:
extracting features of the user information and the thesis information marked by the user to obtain a user identifier and evaluation label data of the thesis;
inputting the thesis information into a training model of Doc2Vec to obtain document vector data of the thesis, wherein the document vector data comprises the identification of the thesis, the subject of the thesis and professional terms in the thesis;
and determining a user-paper relation list according to the label information of the paper label marked by the user, wherein the user-paper relation list comprises a user identifier, a paper identifier, a marking time and a label category.
In some embodiments, the inputting the thesis information into the training model of Doc2Vec to obtain document vector data of a thesis includes:
acquiring thesis information marked by a user;
converting the paper information into a plurality of vectors;
concatenating the plurality of vectors into a sentence vector;
and determining the next word vector according to the sentence vector and the corresponding context information until the document vector data of the thesis is obtained.
In some embodiments, the training a neural network according to the user feature data, the paper feature data, and the score data to obtain an initial recommendation model includes:
learning the linear relation between the user and the paper through a temporal perception matrix decomposition model;
learning the nonlinear relation between the user and the thesis through a multilayer perceptron;
and determining an interactive structure between the user and the thesis according to the learning results of the linear relation and the nonlinear relation.
In some embodiments, the learning of the linear relationship between the user and the paper through the temporal perception matrix decomposition model includes:
taking a user identification and a paper marked by a user as input features, and converting the input features into two-dimensional sparse vectors with single hot spot codes;
mapping the sparse vector to a fully connected layer of dense vectors by an embedding layer;
mapping the prediction scores of the potential vectors through a neural synergy filter layer;
the predicted scores of the papers are obtained through the hidden layer and the output layer.
In some embodiments, the learning of the non-linear relationship between the user and the paper through the multi-layered perceptron includes:
for each user, obtaining a binary vector of the user through single hotspot coding;
integrating the binary vectors of all users into an embedded matrix;
embedding the extracted thesis feature data and the embedded matrix into an input layer; the user feature data and the paper feature data comprise interactive features, attribute features and text features;
and embedding the interactive features, the attribute features and the text features as vectors, inputting the vectors into a multi-layer perceptron, merging the input low-dimensional dense embedded vectors, and determining the nonlinear relation between the user and the thesis.
In some embodiments, the dynamically weighting the initial recommendation model by a temporal perceptual function to obtain a target recommendation model includes:
carrying out dynamic weighting processing on the initial recommendation model through four temporal perception functions;
the four temporal perception functions comprise a linear temporal perception function, a logic temporal perception function, an exponential temporal perception function and an Ebinghauss temporal perception function.
In some embodiments, determining a recommended paper according to the target recommendation model includes:
determining a prediction probability through the target recommendation model, wherein the prediction probability represents the interest probability of the user in the paper which is not labeled;
constructing a paper recommendation list according to the prediction probability of each paper;
and recommending the papers to the user through a Top-N recommendation algorithm according to the paper recommendation list.
The invention provides a paper recommendation system for temporal perception academic information, which comprises:
the characteristic extraction module is used for extracting the characteristics of the user information and the thesis information to obtain user characteristic data, the thesis characteristic data and grading data;
the training module is used for training the neural network according to the user characteristic data, the thesis characteristic data and the grading data to obtain an initial recommendation model;
the weighting processing module is used for carrying out dynamic weighting processing on the initial recommendation model through a temporal perception function to obtain a target recommendation model;
and the recommending module is used for determining a recommended paper according to the target recommending model.
A third aspect of the invention provides a computer readable storage medium storing a program for execution by a processor to perform the method according to the first aspect of the invention.
The embodiment of the invention obtains user characteristic data, thesis characteristic data and grading data by extracting the characteristics of the user information and the thesis information; training a neural network according to the user characteristic data, the thesis characteristic data and the grading data to obtain an initial recommendation model; carrying out dynamic weighting processing on the initial recommendation model through a temporal perception function to obtain a target recommendation model; according to the target recommendation model, the recommended paper is determined, and the accuracy and the practicability of the recommendation system are improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a flowchart illustrating the overall steps of a paper recommendation method according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating a paper recommendation method according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a neural network training process according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a fusion process of TGMF and MLP according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
To solve the problems in the prior art, an embodiment of the present invention provides a method for recommending a thesis of temporal perception academic information, as shown in fig. 1, the method includes the following steps:
extracting the characteristics of the user information and the thesis information to obtain user characteristic data, the thesis characteristic data and grading data;
training a neural network according to the user characteristic data, the thesis characteristic data and the grading data to obtain an initial recommendation model;
carrying out dynamic weighting processing on the initial recommendation model through a temporal perception function to obtain a target recommendation model;
and determining a recommended paper according to the target recommendation model.
In particular, the invention is mainly directed to the problem of paper recommendation in academic social networks. In the present invention, the characteristic information of the user mainly includes the id of the user and the label marked by the interested paper by the user, wherein the kinds of labels are various, generally the research direction or interest of the user at present, and the label without the label is no-tag. The paper characteristics include the id of the paper, publication journal, publication time, and text information (i.e., linking the title and abstract of the paper). The characteristic information can well express the research interest of the user and is more suitable for expressing the relation between the user and the paper, so that the research of paper recommendation in an academic social network can be better carried out.
As shown in fig. 2, in the embodiment of the present invention, feature extraction is performed on the user and the relevant information of the thesis, so as to obtain three parts of data, namely "user", "thesis" and "score"; wherein, the 'score' refers to a label which is marked on a paper which is interested by the user; then, dynamically weighting the user, the paper and the score by using four different temporal perception functions respectively, and inputting the weighted values into a neural network for training to obtain an output result; and finally, recommending the thesis according to the output result.
In the invention, firstly, from the user, the user may have different interests and research directions in different time ranges; from a paper perspective, the reference value of a paper is generally inversely proportional to time, i.e., the longer the paper is published, the lower its value (since some classical algorithms are special cases, the present invention does not take it into account). In addition, the data set used in the present invention includes the interactive behavior between the user and the paper, i.e. the label printed on the paper by the user, so as to indicate that the user is interested in the paper labeled by the user, and the time of label printing is also included herein, so that the present invention should be taken into consideration.
In summary, in the invention, the temporal perception function is mainly used for more accurately recommending papers which may be interested by the user but have not been labeled by the user, the temporal perception function is used as a time weight, and then the dynamic weighting is carried out on the user and the paper vector for training, so that the obtained result is more accurate and effective than an algorithm without considering the time factor.
In some embodiments, the extracting features of the user information and the thesis information to obtain user feature data, thesis feature data, and score data includes:
extracting features of the user information and the thesis information marked by the user to obtain a user identifier and evaluation label data of the thesis;
inputting the thesis information into a training model of Doc2Vec to obtain document vector data of the thesis, wherein the document vector data comprises the identification of the thesis, the subject of the thesis and professional terms in the thesis;
and determining a user-paper relation list according to the label information of the paper label marked by the user, wherein the user-paper relation list comprises a user identifier, a paper identifier, a marking time and a label category.
Specifically, the user feature extraction in the embodiment of the present invention is to obtain the user id and the label printed by the user to the paper according to the user and the paper marked by the user; the paper feature extraction is to assign the extracted 'paper' data documents to a training model of Doc2Vec to obtain vectors for representing different documents, namely to obtain id, subject and professional terms of each paper; where Doc2Vec is an unsupervised algorithm that can learn fixed-length feature representations from variable-length text segments (e.g., sentences, paragraphs, and documents). Compared with Word2Vec, it overcomes the disadvantages of the bag-of-words model: the sequence of the words is lost, the semantics of the words are ignored, and the problem of data sparsity is solved, so that the accuracy is improved.
Specifically, the embodiment of the invention extracts the user-paper relationship. According to the label printed on the paper by the user, a user-paper relation list, namely a 'grading' list is obtained, wherein the user-paper relation list comprises four kinds of characteristic information, namely the user id, the paper id, the label printing time t and the label type tag. Then, presetting a proportion of the first-class thesis and the second-class thesis and sampling; wherein the papers are labeled by the user, namely the papers which are interested by the user; the two types of papers are papers that the user has not marked.
Obtaining a trained model according to a sampling result;
the user-paper data pairs are input into the trained model.
In this embodiment, one type of paper and two types of paper are sampled at a 1:4 ratio. One piece of paper is extracted from a set of papers of the same type, and the piece of paper and a labeled user form a user-paper pair of the same type; then four parts are extracted from the second-class thesis, and the four parts and the users in the previous-class user-thesis pair respectively form four second-class user-thesis pairs; performing model training by using a first-class user-paper pair and four second-class user-paper pairs to obtain a trained model; then repeating the first and second paper extraction steps, and inputting the trained model.
In some embodiments, the inputting the thesis information into the training model of Doc2Vec to obtain document vector data of a thesis includes:
acquiring thesis information marked by a user;
converting the paper information into a plurality of vectors;
concatenating the plurality of vectors into a sentence vector;
and determining the next word vector according to the sentence vector and the corresponding context information until the document vector data of the thesis is obtained.
Specifically, the method for extracting the document theme by using Doc2Vec in the embodiment of the invention mainly comprises the following steps:
firstly, finding out a paper marked by a user, namely a paper with a label;
second, the document is converted into a vector. Wherein the paragraph vector is inferred by fixing the word vector and training a new sentence vector until convergence;
finally, a plurality of word vectors are concatenated into a sentence vector, and then the next word vector is predicted according to a given context.
In some embodiments, the training a neural network according to the user feature data, the paper feature data, and the score data to obtain an initial recommendation model includes:
learning the linear relation between the user and the paper through a temporal perception matrix decomposition model;
learning the nonlinear relation between the user and the thesis through a multilayer perceptron;
and determining an interactive structure between the user and the thesis according to the learning results of the linear relation and the nonlinear relation.
Specifically, the neural network of the embodiment of the present invention includes a neural Matrix decomposition model ncf (neural probabilistic filtering) that integrates TGMF (Time-aware Generalized Matrix Factorization) and MLP (Multi-Layer persistence), and the TGMF immediate perception Matrix decomposition model is an extension of conventional mf (Matrix Factorization) in the neural network, and considers the influence of temporal attributes, thereby learning the linear relationship between the User and the Item of the paper; MLP is a multi-layered perceptron that learns the potentially non-linear characteristics between users and papers. Thus, the present invention unifies the advantages of linearity of MF and non-linearity of MLP to model user-paper potential structures.
In particular, as shown in FIG. 3, embodiments of the present invention employ a multi-layered representation to model the interaction between users-papers, allowing for a systematic filtered complete neural process. Where the output of one layer serves as the input to the next layer.
The bottom input layer consists of two feature vectors of users and papers, and different users and papers correspond to different feature vectors, so that wide user and paper modeling is supported. The id of the user and the labeled paper of the user are used as input features and are converted into a two-dimensional sparse vector with single hot point code (one-hot). Since this is a common input profile representation, users and papers can be represented according to content profiles, thereby solving the cold start problem of data.
The input layer of the embodiment of the invention is an embedding layer which is a full connected layer for mapping sparse vector representation to dense vector, and the obtained User or Item embedding can be regarded as a potential vector of User or Item in a potential model.
The embedding of the users and the papers of the embodiment of the invention is input into a multilayer Neural network, which is named as Neural CF Layers (NCF Layers for short) and is used for mapping the prediction scores of the potential vectors, the same as the input Layers, and different users and papers correspond to different potential vectors, so that the potential characteristics between the users and the papers can be more widely discovered.
And finally, the Output layer obtains a prediction score Output.
The dimensions in the hidden layer of embodiments of the present invention determine the capabilities of the model.
In summary, the model calculation can be expressed as:
wherein P, Q ∈ R is the latent factor matrix, V, of the user and the paper, respectivelyU、VIIs a feature vector of the user and the paper,
and respectively representing mapping functions of the output layer and the nth neural cooperation filtering layer, wherein n neural cooperation filtering layers are provided.
In some embodiments, the learning of the linear relationship between the user and the paper through the temporal perception matrix decomposition model includes:
taking a user identification and a paper marked by a user as input features, and converting the input features into two-dimensional sparse vectors with single hot spot codes;
mapping the sparse vector to a fully connected layer of dense vectors by an embedding layer;
mapping the prediction scores of the potential vectors through a neural synergy filter layer;
the predicted scores of the papers are obtained through the hidden layer and the output layer.
Specifically, the embodiment of the invention describes a learning process of a TGMF model:
since the input layer hot encodes the id attributes of User and Item, the resulting embedded vector can be viewed as a potential vector for User and Item.
Further, using the potential vector cu、ciRespectively represent PVU、QVIThen, a mapping function of the first neural synergistic filter layer is obtained:
wherein an inner product operation sign is a binary operation taking two matrices of the same dimension and generating another matrix of the same dimension, wherein each element i, j is the product of the elements i, j of the original two matrices, cu、ciAre potential vectors for users and papers.
Then, it is mapped to the output layer, resulting in:
Output＝asT*(cu⊙ci)
where Output represents the Output of the TGMF layer, T is the temporal perceptual function, cu、ciIs a potential vector of users and papers, asIs an activation function, and considering linear modeling here, Sigmoid is adopted as the activation function.
Further, the step of mapping the user and the paper to the output layer further comprises:
since the user's interest may change with the migration of time, the influence of the academic papers may gradually decrease with the change of time. That is, as research contents are deepened, the initial research interests of the scholars may tend not only to be stable but also to be shifted to other research fields. Thus, some of the user's recent behaviors are most likely representative of their recent research interests.
In some embodiments, the learning of the non-linear relationship between the user and the paper through the multi-layered perceptron includes:
for each user, obtaining a binary vector of the user through single hotspot coding;
integrating the binary vectors of all users into an embedded matrix;
embedding the extracted thesis feature data and the embedded matrix into an input layer; the user feature data and the paper feature data comprise interactive features, attribute features and text features;
and embedding the interactive features, the attribute features and the text features as vectors, inputting the vectors into a multi-layer perceptron, merging the input low-dimensional dense embedded vectors, and determining the nonlinear relation between the user and the thesis.
Specifically, the embodiment of the invention describes the learning process of the MLP model. The embodiment of the invention embeds the extracted user characteristics and thesis characteristics into an input layer, and the characteristics can be roughly divided into three types: interactive features, attribute features, text features.
The interactive feature is a User-paper list of papers tagged by the User, and comprises a User _ id, a paper ltem _ id, a timestamp and a tag.
The attribute characteristics comprise information such as a paper type, a journal outlet, a publishing company publisher and the like, but only two attributes of the paper type and the journal outlet are selected in the invention.
Text feature Text is the topic and abstract of a paper, as the topic and abstract generally contain the subject matter of the paper, especially the abstract, including the entire process from problem posed to problem analysis, experimentation and resolution. Based on the method, the obtained paper titles and abstracts are connected, and the subjects of each paper are extracted by adopting a Doc2Vec model.
Further, the selected interactive features User _ id, ltem _ id, timemap, tag, attribute feature type, journal and text feature text are embedded as vectors, then input into a standard multi-layer perceptron MLP to learn the potential nonlinear characteristics between users and papers, and the element inner product is not used to describe the interactive characteristics between potential users and items, but the input low-dimensional dense embedded vectors are merged and then sent into a full-connection layer of a neural network through forward transfer to automatically learn the potential nonlinear relationship between the vectors, thereby providing high flexibility and nonlinear modeling capability for the model. Wherein the feature embedding has the effect of mapping the high-dimensional sparse binary vectors to corresponding low-dimensional dense vectors.
Furthermore, for each user, the embodiment of the invention adopts single hot spot coding to obtain the binary vector c of the useru＝[0，1，...，...，0]∈RMEach user corresponds to a unique single hotspot code, where M represents the number of users in the data set.
Further, all users are represented as an inline matrix
Where D is the dimension of the output vector. The binary vector of user U is converted into a low-dimensional dense embedded vector in the matrix.
In summary, the specific calculation process in MLP is as follows:
Z1＝[Uid；Iid；Jid；Typeid；Text]
……
ZL+1＝aL(WLZL+bL)
wherein, Uid、Iid、Jid、TypeidAnd Text is the user id, paper id, publication journal, paper type and paper topic, respectively, as described above, L is the number of layers of the perceptron, W and b are the weight matrix and the offset of the current layer, respectivelyAmount of the compound (A). a isLIs an activation function, unlike in TGMF, considering being a sparse matrix, and in order to prevent data overfitting, better mine the non-linear relationship between users and papers, here, ReLU is chosen as the activation function.
Further, as shown in fig. 4, by fusing the TGMF and the MLP models in the embodiment of the present invention, it can be understood that the TGMF and the MLP models fused in the embodiment of the present invention are based on that the TGMF and the MLP are respectively independently trained to converge, and then respective output results are obtained, and then the two output results are used as parameter inputs of the TGMF-MLP layer. The TGMF applies a temporal perception function as dynamic weight to measure the influence of a paper and the interest of a user, and a linear kernel is applied to model potential feature interaction; MLP learns the interaction between users and papers from data using a non-linear kernel, and performs embedded learning through the academic features obtained in the feature extraction step. And the two models are respectively learned through the embedding of independent users and paper vectors, and then the two models are fused by connecting the last hidden layer of the two models.
In some embodiments, the dynamically weighting the initial recommendation model by a temporal perceptual function to obtain a target recommendation model includes:
carrying out dynamic weighting processing on the initial recommendation model through four temporal perception functions;
the four temporal perception functions comprise a linear temporal perception function, a logic temporal perception function, an exponential temporal perception function and an Ebinghauss temporal perception function.
Specifically, the embodiment of the present invention adopts 4 temporal perception functions to dynamically weight the model, which are respectively:
(1) linear temporal perceptual function W (t)pT): i.e. the time varies linearly:
where t is the current time, tpIs the interaction time of the user and the item, here the userTime of paper labeling; τ represents a weight coefficient representing a time linear change. The larger the change weight coefficient, the faster the speed of change.
(2) Logical temporal perceptual function W (t)pT): the initial state of function growth is an exponentially growing state, and then the growth reaches a relatively steady state near the end.
Where μ is a weight coefficient.
(3) Exponential temporal perceptual function W (t)pT): the time variation being exponential
Wherein γ is also a weighting factor, and the larger γ is, the faster time t changes.
(4) Obbingos time-wise perception function W (t)pT): german psychologist study of memory change curves of human brain
In some embodiments, determining a recommended paper according to the target recommendation model includes:
determining a prediction probability through the target recommendation model, wherein the prediction probability represents the interest probability of the user in the paper which is not labeled;
constructing a paper recommendation list according to the prediction probability of each paper;
and recommending the papers to the user through a Top-N recommendation algorithm according to the paper recommendation list.
Specifically, the embodiment of the invention outputs the prediction result yui，yuiIndicating that the user is interested in papers that he has not taggedProbability, i.e. predictive probability yuiIt can be expressed as:
wherein the content of the first and second substances,
is the output of the TGMF module, learns the interaction between the user and the paper using a linear kernel,
is the output of the MLP module, learns the non-linear relationship between the user and the paper. W (t) is a dynamically weighted temporal perceptual weight.
For more accurate prediction, the model is trained through a Loss function Loss, and the Loss function Loss formula is as follows:
where S denotes the total training set, i.e. all User-paper pairs, y, of User and paper ItemuiE {0, 1}, representing the relationship between the user and the paper in the academic social network and the prediction probability
Correspondingly, 1 indicates that the user tagged the paper, and 0 indicates that no tagging was made.
Finally, according to the prediction probability
A list of recommendations of a paper is constructed,
the larger the value of (A), the more the papers are ranked, and then K papers which are ranked first are recommended to the user as a recommendation result, wherein K can be set with different values according to needs.
The embodiment of the invention also provides a system for recommending a thesis of the tense perception academic information, which comprises the following steps:
the characteristic extraction module is used for extracting the characteristics of the user information and the thesis information to obtain user characteristic data, the thesis characteristic data and grading data;
the training module is used for training the neural network according to the user characteristic data, the thesis characteristic data and the grading data to obtain an initial recommendation model;
the weighting processing module is used for carrying out dynamic weighting processing on the initial recommendation model through a temporal perception function to obtain a target recommendation model;
and the recommending module is used for determining a recommended paper according to the target recommending model.
An embodiment of the present invention further provides a computer-readable storage medium, where the storage medium stores a program, and the program is executed by a processor to implement the method shown in fig. 1.
The embodiment of the invention also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and executed by the processor to cause the computer device to perform the method illustrated in fig. 1.
Compared with the prior art, the method and the device have the advantages that the academic characteristics of the user and the thesis are sensed through the tense; and inputting the model, obtaining an output result after dynamic weighting, and recommending a thesis according to the output result. The recommended results are compared using four temporal perception functions, and our model is evaluated using multiple evaluation indices. The method can effectively consider the characteristics of the thesis and the change of the interest of the scholars, solve the problems of cold start, sparseness and the like of data, filter the expired thesis and effectively control the timeliness problem of the data, thereby greatly improving the accuracy of the recommendation of the thesis.
In alternative embodiments, the functions/acts noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved. Furthermore, the embodiments presented and described in the flow charts of the present invention are provided by way of example in order to provide a more thorough understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of larger operations are performed independently.
Furthermore, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or software module, or one or more functions and/or features may be implemented in a separate physical device or software module. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be understood within the ordinary skill of an engineer, given the nature, function, and internal relationship of the modules. Accordingly, those skilled in the art can, using ordinary skill, practice the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative of and not intended to limit the scope of the invention, which is defined by the appended claims and their full scope of equivalents.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. A method for recommending a thesis of temporal perception academic information is characterized by comprising the following steps:
extracting the characteristics of the user information and the thesis information to obtain user characteristic data, the thesis characteristic data and grading data;
training a neural network according to the user characteristic data, the thesis characteristic data and the grading data to obtain an initial recommendation model;
carrying out dynamic weighting processing on the initial recommendation model through a temporal perception function to obtain a target recommendation model;
and determining a recommended paper according to the target recommendation model.
2. The method of claim 1, wherein the extracting the features of the user information and the thesis information to obtain the user feature data, the thesis feature data and the score data comprises:
extracting features of the user information and the thesis information marked by the user to obtain a user identifier and evaluation label data of the thesis;
inputting the thesis information into a training model of Doc2Vec to obtain document vector data of the thesis, wherein the document vector data comprises the identification of the thesis, the subject of the thesis and professional terms in the thesis;
and determining a user-paper relation list according to the label information of the paper label marked by the user, wherein the user-paper relation list comprises a user identifier, a paper identifier, a marking time and a label category.
3. A paper recommendation method according to claim 1, wherein the step of inputting the paper information into a training model of Doc2Vec to obtain document vector data of a paper comprises:
acquiring thesis information marked by a user;
converting the paper information into a plurality of vectors;
concatenating the plurality of vectors into a sentence vector;
and determining the next word vector according to the sentence vector and the corresponding context information until the document vector data of the thesis is obtained.
4. The paper recommendation method of temporal perception academic information according to claim 1, wherein the training of the neural network according to the user feature data, the paper feature data and the score data to obtain an initial recommendation model comprises:
learning the linear relation between the user and the paper through a temporal perception matrix decomposition model;
learning the nonlinear relation between the user and the thesis through a multilayer perceptron;
and determining an interactive structure between the user and the thesis according to the learning results of the linear relation and the nonlinear relation.
5. The method of claim 4, wherein learning the linear relationship between the user and the paper through the temporal perception matrix decomposition model comprises:
taking a user identification and a paper marked by a user as input features, and converting the input features into two-dimensional sparse vectors with single hot spot codes;
mapping the sparse vector to a fully connected layer of dense vectors by an embedding layer;
mapping the prediction scores of the potential vectors through a neural synergy filter layer;
the predicted scores of the papers are obtained through the hidden layer and the output layer.
6. The method of claim 4, wherein learning the non-linear relationship between the user and the paper through the multi-layered perceptron comprises:
for each user, obtaining a binary vector of the user through single hotspot coding;
integrating the binary vectors of all users into an embedded matrix;
embedding the extracted thesis feature data and the embedded matrix into an input layer; the user feature data and the paper feature data comprise interactive features, attribute features and text features;
and embedding the interactive features, the attribute features and the text features as vectors, inputting the vectors into a multi-layer perceptron, merging the input low-dimensional dense embedded vectors, and determining the nonlinear relation between the user and the thesis.
7. The method of claim 1, wherein the dynamically weighting the initial recommendation model by a temporal perception function to obtain a target recommendation model comprises:
carrying out dynamic weighting processing on the initial recommendation model through four temporal perception functions;
the four temporal perception functions comprise a linear temporal perception function, a logic temporal perception function, an exponential temporal perception function and an Ebinghauss temporal perception function.
8. A paper recommendation method according to claim 1, wherein determining a recommended paper according to the target recommendation model comprises:
determining a prediction probability through the target recommendation model, wherein the prediction probability represents the interest probability of the user in the paper which is not labeled;
constructing a paper recommendation list according to the prediction probability of each paper;
and recommending the papers to the user through a Top-N recommendation algorithm according to the paper recommendation list.
9. A system for recommending thesis based on temporal perception academic information, comprising:
the characteristic extraction module is used for extracting the characteristics of the user information and the thesis information to obtain user characteristic data, the thesis characteristic data and grading data;
the training module is used for training the neural network according to the user characteristic data, the thesis characteristic data and the grading data to obtain an initial recommendation model;
the weighting processing module is used for carrying out dynamic weighting processing on the initial recommendation model through a temporal perception function to obtain a target recommendation model;
and the recommending module is used for determining a recommended paper according to the target recommending model.
10. A computer-readable storage medium, characterized in that the storage medium stores a program, which is executed by a processor to implement the method according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011050718.3A CN112214687B (en) | 2020-09-29 | 2020-09-29 | Paper recommendation method, system and medium for tense perception academic information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011050718.3A CN112214687B (en) | 2020-09-29 | 2020-09-29 | Paper recommendation method, system and medium for tense perception academic information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112214687A true CN112214687A (en) | 2021-01-12 |
| CN112214687B CN112214687B (en) | 2023-06-06 |
Family
ID=74052459
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011050718.3A Active CN112214687B (en) | 2020-09-29 | 2020-09-29 | Paper recommendation method, system and medium for tense perception academic information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112214687B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113343125A (en) * | 2021-06-30 | 2021-09-03 | 南京大学 | Academic-precision-recommendation-oriented heterogeneous scientific research information integration method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180211174A1 (en) * | 2017-01-24 | 2018-07-26 | International Business Machines Corporation | System for evaluating journal articles |
| CN109785062A (en) * | 2019-01-10 | 2019-05-21 | 电子科技大学 | A kind of hybrid neural networks recommender system based on collaborative filtering model |
| CN111581401A (en) * | 2020-05-06 | 2020-08-25 | 西安交通大学 | Local citation recommendation system and method based on depth correlation matching |
-
2020
- 2020-09-29 CN CN202011050718.3A patent/CN112214687B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180211174A1 (en) * | 2017-01-24 | 2018-07-26 | International Business Machines Corporation | System for evaluating journal articles |
| CN109785062A (en) * | 2019-01-10 | 2019-05-21 | 电子科技大学 | A kind of hybrid neural networks recommender system based on collaborative filtering model |
| CN111581401A (en) * | 2020-05-06 | 2020-08-25 | 西安交通大学 | Local citation recommendation system and method based on depth correlation matching |
Non-Patent Citations (1)
| Title |
|---|
| 唐浩;刘柏嵩;刘晓玲;黄伟明;: "基于协同知识图谱特征学习的论文推荐方法", 计算机工程 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113343125A (en) * | 2021-06-30 | 2021-09-03 | 南京大学 | Academic-precision-recommendation-oriented heterogeneous scientific research information integration method and system |
| CN113343125B (en) * | 2021-06-30 | 2023-08-22 | 南京大学 | Academic accurate recommendation-oriented heterogeneous scientific research information integration method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112214687B (en) | 2023-06-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hjørland | Data (with big data and database semantics) | |
| Benabderrahmane et al. | On the predictive analysis of behavioral massive job data using embedded clustering and deep recurrent neural networks | |
| Kulkarni et al. | Deep learning for NLP | |
| CN110083836B (en) | Key evidence extraction method for text prediction result | |
| Ma et al. | MMM: multi-source multi-net micro-video recommendation with clustered hidden item representation learning | |
| Ji et al. | Cross-domain sentiment classification via a bifurcated-LSTM | |
| Wang et al. | Joint representation learning with ratings and reviews for recommendation | |
| Ibrahim et al. | An intelligent hybrid neural collaborative filtering approach for true recommendations | |
| CN114003726B (en) | Subspace embedding-based academic thesis difference analysis method | |
| Firdaus et al. | Sentiment guided aspect conditioned dialogue generation in a multimodal system | |
| Ghorbanali et al. | Exploiting bi-directional deep neural networks for multi-domain sentiment analysis using capsule network | |
| Li et al. | Novel recommendation system for tourist spots based on hierarchical sampling statistics and SVD++ | |
| CN112214687A (en) | Paper recommendation method, system and medium for temporal perception academic information | |
| Issam et al. | Financial sentiment analysis of tweets based on deep learning approach | |
| Achilles et al. | Using Surface and Semantic Features for Detecting Early Signs of Self-Harm in Social Media Postings. | |
| Fosset et al. | Docent: A content-based recommendation system to discover contemporary art | |
| Al Mubasher et al. | A novel neural network-based recommender system for drug recommendation | |
| Wen et al. | Hierarchical text interaction for rating prediction | |
| CN115269984A (en) | Professional information recommendation method and system | |
| Lu | Design of a music recommendation model on the basis of multilayer attention representation | |
| Ochoa-Luna et al. | Word embeddings and deep learning for spanish twitter sentiment analysis | |
| Oleinik | Detection of opinion communities with the help of chance-corrected measures of agreement | |
| Fu et al. | Sequence-based recommendation with bidirectional lstm network | |
| Martin et al. | Fusion-based Representation Learning Model for Multimode User-generated Social Network Content | |
| Habbat et al. | Analyzing Booking's Comments Using Stacking Ensemble Deep Learning Model and Neural Topic Model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |