CN112287222A - Content recommendation method based on heterogeneous feature depth residual error network - Google Patents
Content recommendation method based on heterogeneous feature depth residual error network Download PDFInfo
- Publication number
- CN112287222A CN112287222A CN202011180629.0A CN202011180629A CN112287222A CN 112287222 A CN112287222 A CN 112287222A CN 202011180629 A CN202011180629 A CN 202011180629A CN 112287222 A CN112287222 A CN 112287222A
- Authority
- CN
- China
- Prior art keywords
- data
- content
- residual error
- error network
- depth residual
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000012545 processing Methods 0.000 claims abstract description 22
- 238000006243 chemical reaction Methods 0.000 claims abstract description 17
- 238000004422 calculation algorithm Methods 0.000 claims description 48
- 239000011159 matrix material Substances 0.000 claims description 17
- 230000009467 reduction Effects 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 6
- 238000013501 data transformation Methods 0.000 claims 1
- 238000004364 calculation method Methods 0.000 abstract description 7
- 239000013598 vector Substances 0.000 description 25
- 238000010586 diagram Methods 0.000 description 11
- 238000012549 training Methods 0.000 description 9
- 238000000513 principal component analysis Methods 0.000 description 7
- 238000001914 filtration Methods 0.000 description 6
- 238000000354 decomposition reaction Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 3
- 238000005065 mining Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000001965 increasing effect Effects 0.000 description 2
- 238000002156 mixing Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 210000004205 output neuron Anatomy 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000002195 synergetic effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
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
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses a content recommendation method based on a heterogeneous characteristic depth residual error network, which comprises the following steps: acquiring source data of content to be recommended, and performing data conversion processing on the source data to obtain weighted mixed embedded data; obtaining prediction scoring data according to the weighted mixed embedded data and the depth residual error network model; and obtaining a recommendation result of the content according to the prediction scoring data. According to the embodiment of the invention, the source data of the field content is processed, the processed data is input into the residual error network model to obtain the prediction scoring data, and then the field content is accurately recommended according to the prediction scoring data, so that the calculation method is high in efficiency and low in resource occupancy rate, and meanwhile, the diversified data can avoid cold start when a new user is recommended.
Description
Technical Field
The invention relates to the technical field of communication, in particular to a content recommendation method based on a heterogeneous characteristic deep residual error network.
Background
With the arrival of the big data era, people have great demands on content recommendation of fields such as movies, articles, medical treatment and the like in the big data era, in order to make good recommendation of the content of the fields, a plurality of methods are developed in the prior art, but the prior art method has uncertainty of decomposition and proof steps, a decomposed low-dimensional matrix increases the time complexity of an algorithm, so that the calculation consumption is high, the efficiency is low, the cold start problem can occur when a new user is met, and the prior art method directly performs operation on sparse data, so that the resource occupancy rate is high, and the algorithm efficiency is low.
Thus, there is still a need for improvement and development of the prior art.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a content recommendation method based on a heterogeneous feature deep residual error network, aiming at solving the problem of cold start when a new user is encountered in the prior art, and the problems that the prior art method directly performs operation on sparse data, the resource occupancy rate is high, and the calculation method efficiency is low.
The technical scheme adopted by the invention for solving the problems is as follows:
in a first aspect, an embodiment of the present invention provides a content recommendation method based on a heterogeneous feature depth residual error network, where the method includes:
acquiring source data of content to be recommended, and performing data conversion processing on the source data to obtain weighted mixed embedded data;
obtaining prediction scoring data according to the weighted mixed embedded data and the depth residual error network model;
and obtaining a recommendation result of the content according to the prediction scoring data.
In one implementation, the performing data conversion processing on the source data to obtain weighted mixed embedded data includes:
acquiring collaborative relationship data of a user and a project;
extracting a characteristic value of the source data of the content to be recommended to obtain characteristic value data;
embedding the characteristic value data and the collaborative relationship data of the user and the project based on content to obtain embedded data;
and generating a first triplet loss trainer, and inputting the embedded data into the first triplet loss trainer to obtain weighted mixed embedded data.
In one implementation manner, the extracting a feature value of source data of the content to be recommended to obtain feature value data includes:
according to the mapping matrix, carrying out linear dimensionality reduction on the source data of the content to be recommended to obtain dimensionality reduction matrix data;
and extracting the characteristic value of the dimension reduction matrix data to obtain characteristic value data.
In one implementation, the obtaining prediction score data according to the weighted hybrid embedded data and the depth residual network model includes:
according to the weighted mixed embedded data, extracting heterogeneous characteristic data of the weighted mixed embedded data;
linearly splicing the weighted mixed embedded data and the heterogeneous characteristic data to obtain deep residual error network input data;
and inputting the depth residual error network input data into a depth residual error network model to obtain prediction scoring data.
In one implementation, the depth residual error network model is generated by:
acquiring input sample data and output sample data;
generating a second triplet loss trainer, and generating a network model according to the second triplet loss trainer;
inputting the input sample data into a network model to obtain network model output data;
and when the average absolute error rate of the network model output data and the output sample data is less than a preset value, stopping iteration to obtain a deep residual error network model.
In one implementation, the obtaining the recommendation result of the content according to the prediction scoring data includes:
obtaining a Baseline algorithm result according to a Baseline algorithm;
and obtaining a recommendation result of the content according to the baseline algorithm result and the prediction scoring data.
In one implementation, the obtaining a recommendation of content based on the baseline algorithm result and the predictive scoring data includes:
obtaining baseline scoring data, user deviation data and project deviation data according to the baseline algorithm result;
adjusting the prediction scoring data according to the baseline scoring data, the user deviation data and the project deviation data to obtain target prediction scoring data;
and obtaining a recommendation result of the content according to the target prediction scoring data.
In one implementation, the obtaining a recommendation result of a content according to the target prediction scoring data further includes:
and when a new user and a new project are introduced, the baseline scoring data is used as target prediction scoring data to obtain a recommendation result of the content.
In a second aspect, an embodiment of the present invention further provides a content recommendation device based on a heterogeneous feature deep residual error network, where the device includes:
the device comprises a source data acquisition unit of the content to be recommended, a source data acquisition unit of the content to be recommended and a weighted mixed embedded data acquisition unit, wherein the source data acquisition unit is used for acquiring source data of the content to be recommended and carrying out data conversion processing on the source data to obtain weighted mixed embedded data;
the prediction scoring data acquisition unit is used for acquiring prediction scoring data according to the weighted mixed embedded data and the depth residual error network model;
and the content recommendation result acquisition unit is used for acquiring a recommendation result of the content according to the prediction scoring data.
In a third aspect, an embodiment of the present invention further provides an intelligent terminal, including a memory, and one or more programs, where the one or more programs are stored in the memory, and configured to be executed by one or more processors, where the one or more programs include a program for executing the content recommendation method based on the heterogeneous feature depth residual error network according to any one of the above items.
In a fourth aspect, the present invention also provides a non-transitory computer-readable storage medium, wherein instructions, when executed by a processor of an electronic device, enable the electronic device to perform a content recommendation method based on a heterogeneous feature depth residual network as described in any one of the above.
The invention has the beneficial effects that: the method comprises the steps of firstly obtaining source data of contents to be recommended in a certain field, and carrying out data conversion processing on the source data to obtain weighted mixed embedded data; then, content prediction scoring data of the field is obtained according to the weighted mixed embedded data and the depth residual error network model; finally, according to the predicted grading data, the final content recommendation result of the field can be given. Therefore, the method and the device for recommending the field content have the advantages that the source data of the field content are processed, the processed data are input into the residual error network model, the forecast scoring data are obtained, then the field content is accurately recommended according to the forecast scoring data, the computing method is high in efficiency and low in resource occupancy rate, and meanwhile the diversified data can avoid cold start when a new user is recommended.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of a content recommendation method based on a heterogeneous feature deep residual error network according to an embodiment of the present invention
Fig. 2 is a diagram of an overall network structure of the HRN according to the embodiment of the present invention
FIG. 3 is a diagram of a heterogeneous feature extraction architecture according to an embodiment of the present invention
FIG. 4 is a graph of RMSE variation trend of the left example of the present invention
FIG. 4 is a LOSS trend chart of the vector experiment result of the embodiment of the present invention
FIG. 5 is a MAE trend graph of vector experiment results according to an embodiment of the present invention
FIG. 6 is a variation diagram of the deep residual error network MAE according to the embodiment of the present invention
FIG. 7 is a diagram of the variation of the depth residual error network RMSE according to the embodiment of the present invention
FIG. 8 is a variation diagram of deep residual network MAP according to an embodiment of the present invention
FIG. 9 is a graph of the change of the depth residual error network MRR according to the embodiment of the present invention
FIG. 10 is a graph of the change of the depth residual network NDCG of the embodiment of the present invention
Fig. 11 is a schematic block diagram of a content recommendation device based on a heterogeneous feature depth residual error network according to an embodiment of the present invention.
Fig. 12 is a schematic block diagram of an internal structure of an intelligent terminal according to an embodiment of the present invention.
Detailed Description
The invention discloses a content recommendation method based on a heterogeneous characteristic deep residual error network, which is further described in detail below by referring to the attached drawings and embodiments in order to make the purposes, technical schemes and effects of the invention clearer and clearer. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
In the prior art, the content recommendation method has the cold start problem when a new user is met, and the prior art method directly performs operation on sparse data, so that the resource occupancy rate is high and the algorithm efficiency is low.
In order to solve the problems of the prior art, the embodiment provides a content recommendation method based on a heterogeneous feature depth residual error network, which includes the steps of obtaining source data of content to be recommended, and after user data, item data and user item score data are obtained, performing data conversion processing on the source data to facilitate subsequent processing on the data to obtain weighted mixed embedded data; then extracting features according to the weighted mixed embedded data, vectorizing the features, converting the vectorized features into an embedded format required by a recommendation algorithm, and inputting the weighted mixed embedded data and the extracted data into a depth residual error network model to obtain predicted scoring data; and finally, obtaining a recommendation result of the content according to the prediction scoring data. The core of the deep residual error network in the embodiment of the invention is HRN, and the HRN is a Chinese abbreviation of a hybrid residual error network. The HRN basic Network has a four-layer residual Network structure, where h-4 denotes a basic Network depth of 4, and is formed by stacking four layers of resnets. The network receives the vector and the characteristic value as input, the vector and the characteristic value enter a Dense layer (full connection layer) respectively, the Dense layer adopts tanh (hyperbolic tangent function) as an activation function, output neurons of the two Dense layers are spliced, and the spliced output neurons are input into the mixed residual error network.
In order to better extract the features of different levels of data, a deeper network is generally designed, and in addition, the more abstract the extracted features of the deeper network are, the more the hidden features of the data can be captured. However, simply relying on increasing the number of layers of the neural network not only fails to achieve the objective but also causes gradient explosions. In order to solve the problem, a ResNet (residual error network) is introduced into the system, a Dropout (discarding the neural network unit from the network according to a certain probability) layer is added into the network to prevent overfitting, and a Batch _ Normalization layer is introduced to carry out Normalization operation on input data and data of the middle layer, so that the network can be ensured to adopt random gradient descent in back propagation, and the network can be converged.
For example, the embodiment of the present invention may be applied to various recommendation fields, and articles, even medical treatment, etc. may be provided in addition to movies, for example, taking a recommended movie as an example, first, items such as a plurality of movie rating data and movie watching user data need to be acquired, then, data conversion processing is performed on the rating data of the movies and the movie watching user data, weighted mixed embedded data related to movie content may be acquired after data conversion processing, feature values are extracted according to the weighted mixed embedded data, for example, viewing rating data of N users are extracted according to the rating data of movies, data conversion processing is performed on the data to acquire weighted mixed embedded data, feature value data of the weighted mixed embedded data is extracted, the feature value data and the weighted mixed embedded data are input to a depth residual error network, and predicted rating data may be acquired, that is, it can be obtained that the model a user and the model B user have many similar viewing preferences, and finally, according to the predicted rating data, if the viewing items of the model B user are many, the item content with the high viewing rating of the model B user that the model a user does not see can be recommended to the model a user.
Exemplary method
The embodiment provides a content recommendation method based on a heterogeneous feature depth residual error network, and the method can be applied to an IT intelligent terminal. As shown in fig. 1 in detail, the method includes:
and S100, acquiring source data of the content to be recommended, and performing data conversion processing on the source data to obtain weighted mixed embedded data.
Specifically, source data of the content to be recommended is acquired, and after user data, project data and user project score data are acquired, data conversion processing is performed on the source data to facilitate subsequent data processing, so that weighted mixed embedded data are obtained. For example, according to rating data of a movie, viewing rating data of N users are extracted, the data are subjected to data conversion processing to obtain weighted mixed embedded data, and the obtained data are prepared for subsequent HRN processing. The HRN overall network structure diagram is shown in fig. 2, and the algorithm flow is as follows: 1) extracting and vectorizing data features; 2) constructing a content-based embedding; 3) enhancing content-based embedding by utilizing a user project collaborative relationship, and constructing weighted mixed embedding; 4) constructing heterogeneous features, splicing the heterogeneous features and weighting, mixing and embedding; 5) constructing a depth residual error prediction network; 6) training a residual error network by using heterogeneous characteristics and weighted mixed embedding; 7) and obtaining the final prediction result.
In order to obtain the weighted mixed embedded data, the data conversion processing is performed on the source data, and obtaining the weighted mixed embedded data includes the following steps:
s101: acquiring collaborative relationship data of a user and a project;
s102: extracting a characteristic value of the source data of the content to be recommended to obtain characteristic value data;
s103: embedding the characteristic value data and the collaborative relationship data of the user and the project based on content to obtain embedded data;
s104: and generating a first triplet loss trainer, and inputting the embedded data into the first triplet loss trainer to obtain weighted mixed embedded data.
Specifically, collaborative relationship data of users and items, such as a user id index, an item id index, a user vector, an item vector, user recurrence times, that is, each user scores each item, item recurrence times, that is, each item receives a score from each user, a content-based user vector, a content-based item vector, a collaborative filtering-based user vector, and a collaborative filtering-based item vector, are obtained. And then extracting the characteristic value of the source data of the content to be recommended to obtain characteristic value data. Common statistical-based feature extraction methods include: the maxima, minima, standard deviations/variances, etc. of the data set are extracted. And then, carrying out content-based embedding construction on the characteristic value data and the collaborative relationship data of the user and the project to obtain embedded data. That is, the score received by each item from each user, the user vector based on the content, the item vector based on the content, the user vector based on the collaborative filtering, the item vector based on the collaborative filtering, the maximum value, the average value, the standard deviation/variance of the data set, and the like are embedded to obtain embedded data. Finally, a first triplet loss trainer is generated and the embedded data is input to the first triplet loss trainer, the triplet loss being a triplet loss function. The first triple Loss module is responsible for enhancing the generated content-based embedding by utilizing the user project system relationship, training the generated content-based embedding into weighted mixed embedding, and then generating a module in the displayed content-based embedding vector, and is responsible for converting the input user and project data into the content-based embedding vector.
In order to obtain suitable characteristic value data, the extracting the characteristic value of the source data of the content to be recommended, and obtaining the characteristic value data includes the following operations: according to the mapping matrix, carrying out linear dimensionality reduction on the source data of the content to be recommended to obtain dimensionality reduction matrix data; and extracting the characteristic value of the dimension reduction matrix data to obtain characteristic value data.
Specifically, according to a mapping matrix, linear dimensionality reduction processing is performed on the source data of the content to be recommended, the dimensionality reduction matrix adopts PCA (principal component analysis), the PCA uses an orthogonal transformation matrix as the mapping matrix, initial data is transformed into a group of representations which are linearly independent of each dimensionality, the representations are used for extracting main characteristic components of the data, the purpose that the overall data is expressed by data with few characteristics is achieved, and the method is a linear dimensionality reduction method. The deep residual error network uses Principal Component Analysis (PCA) (principal Component analysis) to perform dimensionality reduction on the high-dimensional matrix embedding after characterization, maps the high-dimensional matrix embedding into a low-dimensional space, converts the high-dimensional sparse embedding into low-dimensional dense embedding, improves the space utilization rate, and can solve the problems of high memory occupation and long calculation time.
The embodiment provides a content recommendation method based on a heterogeneous feature depth residual error network, and the method can be applied to an IT intelligent terminal. As shown in fig. 1 in detail, the method includes:
s200: and obtaining prediction scoring data according to the weighted mixed embedded data and the depth residual error network model.
Specifically, after obtaining the weighted mixed embedded data, the embodiment of the present invention inputs the data into a deep residual error network model, where the core of the deep residual error network model is HRN, and the HRN is designed according to the following criteria: 1) in order to reduce the influence of cold start on the recommendation result, the HRN designs a triple Loss trainer, trains out and considers the content-based and collaborative filtering-based weighted mixed embedding as the input of a prediction network. 2) Second, because of the large number of users and items, the number of items that interact with a user is only a very small fraction. In order to avoid direct operation on an extremely sparse user-to-item evaluation matrix, the HRN utilizes PCA to reduce the dimension of embedding, uses triple Loss to enhance content-based embedding by using a user-item synergistic relationship, converts original user and item data into low-dimension data, and simultaneously keeps a good weighted mixed embedding vector of feature mapping. And inputting the weighted mixed embedded data into the depth residual error network model to obtain the prediction evaluation data.
In order to obtain the prediction score data, the obtaining of the prediction score data according to the weighted mixed embedded data and the depth residual error network model comprises the following steps:
s201: according to the weighted mixed embedded data, extracting heterogeneous characteristic data of the weighted mixed embedded data;
s202: linearly splicing the weighted mixed embedded data and the heterogeneous characteristic data to obtain deep residual error network input data;
s203: and inputting the depth residual error network input data into a depth residual error network model to obtain prediction scoring data.
In practice, after the weighted mixed embedded data is obtained, heterogeneous feature data of the weighted mixed embedded data is extracted, and the heterogeneous feature extraction aims at mining implicit relations among different types of features. The heterogeneous information network HIN comprises a directed graph, wherein entities are represented by nodes, and relationships are represented by edges, and the directed graph is particularly similar to a knowledge graph. Inspired by the HIN heterogeneous information network, the system fuses various learned feature vectors into heterogeneous feature vectors in a linear splicing mode, and linearly splices the heterogeneous feature vectors and the learned homogenous vectors (sequences) to serve as input of a deep residual error network. Based on this, we build a heterogeneous feature extraction as shown in fig. 3. Then linearly splicing the weighted mixed embedded data and the heterogeneous characteristic data to form data serving as deep residual error network input data; because the depth residual error network model is well established, the prediction scoring data can be obtained by inputting the depth residual error network input data into the depth residual error network model.
In order to use the depth residual error network model, the depth residual error network model is firstly generated, and the generation mode of the depth residual error network model is as follows: acquiring input sample data and output sample data; generating a second triplet loss trainer, and generating a network model according to the second triplet loss trainer; inputting the input sample data into a network model to obtain network model output data; and when the average absolute error rate of the network model output data and the output sample data is less than a preset value, stopping iteration to obtain a deep residual error network model.
In practice, a network model needs to be obtained by training a model according to a set of real sample data, and therefore, input data and output data need to be obtained, and then a second Triplet Loss trainer is generated, Triplet Loss is a triple Loss function, Triplet Loss is used for learning embedding for face image refinement, embedding is a mode of converting discrete variables into continuous vector representation, and similar images are also similar in an embedding space. Triplet Loss possesses a finer granularity than a traditional trainer. The triple Loss is more suitable for accurate identification, is mainly applied to fine-grained identification problems such as face identification, identity identification, vehicle identification and the like, can greatly reduce the output dimension of a network, and can also obtain better feature embedding. And then inputting input sample data into a network model to obtain network model output data, wherein the output data is obtained by the network model according to the input data, in order to obtain a final depth residual error network model, the average absolute error rate of the output data of the network model and the average absolute error rate of the output sample data are required to be evaluated, when the average absolute error rate of the output data of the network model and the average absolute error rate of the output sample data are smaller than a preset value, the model training is successful, the iteration is stopped, and the obtained depth residual error network model can be used.
After the deep residual network training in the embodiment of the present invention, the variation trend of RMSE (root mean square error) with the increase of epoch (one generation of training) is shown in the left side of fig. 4, the variation trend of LOSS with the increase of epoch (one generation of training) is shown in the right side of fig. 4, and the variation trend of MAE (mean absolute error) with the increase of epoch (one generation of training) is shown in fig. 5.
As the number of iterations increases, the LOSS, RMSE and MAE of the network gradually decrease, and when the epoch reaches 30, the LOSS, RMSE and MAE of the network decrease from the beginning 3.6843,0.8995,0.7122 to 3.6234,0.8465,0.6694, respectively. It can be seen that over 30 iterations, LOSS, RMSE and MAE of the depth residual network all fall to a relatively low level.
As can be seen from fig. 4 left, fig. 4 right, and fig. 5, for each index, the algorithm participating in comparison is roughly divided into three camps. The system comprises a Normal (orbit algorithm), a base (Baseline algorithm) and a CoCluster (clustering algorithm) which are classified into a formation, an SVD (singular value decomposition algorithm) and an SVD + + (singular value decomposition improvement algorithm) which are classified into a formation, a deep residual network based on heterogeneous characteristics, a Hybrid _ ResNet (HRN) algorithm trained by using weighted Hybrid embedding, and a formation formed by Hybrid _ SVD (Hybrid singular value decomposition algorithm) and Hybrid _ SVD + + (Hybrid singular value decomposition improvement algorithm) algorithms which are trained by using SVD and SVD + +, respectively. It can be seen that as the user score increases, each index of each algorithm becomes better, which also indicates that the more the user interacts with the project, the better the algorithm can capture the preference of the user, and better prediction can be made. Through transverse comparison, the three algorithms HRN, HSD and HSD + + trained by the method provided by the system are found to be obviously superior to other algorithms such as SVD and SVD + + algorithms. Secondly, Baseline, CoClustering, Normal algorithms perform poorly. SVD + + as the improved version of SVD, except the first lineup that the system proposed, the performance is the best, each index all improves to some extent compared with SVD.
The three networks, namely HRN, HSD and HSD + +, proposed by the system are built on the basis of the same weighted mixed embedding which integrates content-based characteristics and collaborative filtering characteristics, and the difference is that after the Hybrid _ ResNet is compared, heterogeneous characteristics are added for mining implicit characteristics of user projects, the heterogeneous characteristics are input into a deep residual error network, and compared with a common deep error network, the learning efficiency of the residual error network is improved, the deeper network structure is also beneficial to learning more abstract characteristics, and the network performance cannot be influenced, which is also the reason for selecting the residual error network by the system. And the HSD + +, on the basis of the weighted mixing embedding, respectively train the network by using SVD and SVD + + algorithms. From the performance comparison result, the indexes of the HRN are better than those of the HRN, and the fact that mining of the user project implicit characteristics by using the deep residual error network is really helpful for improving the network prediction performance is proved.
The more interactive data a user interacts with a project, the more the recommendation algorithm can learn the user's preferences, and in terms of taking an ML (canonical dataset) (1M) dataset, the more movies the user participates in scoring, the less the prediction scoring error given by the HRN. With the increase of the user-rating-0count user score, fig. 6 shows the variation of MAE (mean absolute error), fig. 7 shows the variation of RMSE (root mean square error), fig. 8 shows the variation of MAP (mean precision average), fig. 9 shows the variation of MRR (mean reciprocal rank), fig. 10 shows the variation of NDCG (normalized breaking cumulative gain), and as the user score data is richer, the algorithm can capture the behavior feature preference of the user more reliably, and make more accurate prediction.
The embodiment provides a content recommendation method based on a heterogeneous feature depth residual error network, and the method can be applied to an IT intelligent terminal. As shown in fig. 1 in detail, the method includes:
s300: and obtaining a recommendation result of the content according to the prediction scoring data.
Specifically, the HRN takes the weighted mixed embedding and the heterogeneous feature vectors as input of a depth residual error network, and obtains prediction score data through training. The HRN fuses feature vectors with different properties into heterogeneous feature vectors, similarity evaluation is carried out on a new user and a new project as well as other users and projects, and content recommendation is made according to the similarity evaluation.
In order to obtain a detailed recommendation result, the obtaining of the recommendation result of the content according to the prediction scoring data comprises the following steps:
s301: obtaining a Baseline algorithm result according to a Baseline algorithm;
s302: and obtaining a recommendation result of the content according to the baseline algorithm result and the prediction scoring data.
In this embodiment, the Baseline algorithm is an algorithm independent from the deep residual error network, before the deep residual error network is constructed, a Baseline algorithm is established, and the final recommendation result is the combination of the result obtained by the Baseline algorithm and the result obtained by the HRN. The introduction of the baseline algorithm allows a network provided by the system to have a comparison baseline, and is beneficial to the improvement of the effect of a subsequent observation algorithm. And finally, correcting the predicted scoring data according to the baseline algorithm result to obtain final scoring data, and obtaining a recommendation result of the field content according to the final scoring data.
In specific implementation, the obtaining of the recommendation result of the content according to the baseline algorithm result and the prediction score data includes the following operations: obtaining baseline scoring data, user deviation data and project deviation data according to the baseline algorithm result; adjusting the prediction scoring data according to the baseline scoring data, the user deviation data and the project deviation data to obtain target prediction scoring data; and obtaining a recommendation result of the content according to the target prediction scoring data.
In one embodiment, since the basic idea of the Baseline algorithm is: a baseline is established, introducing user and project biases. For example, the user deviation in the movie domain is the deviation of the score of the viewing habits of each user, and the project deviation in the movie domain is the deviation of the work itself of each movie. Adjusting the prediction scoring data according to the baseline scoring data, the user deviation data and the project deviation data to obtain target prediction scoring data; that is, according to the deviation of the rating of the viewing habits of each user and the deviation of the works of each movie, the prediction rating can be adjusted to obtain target prediction rating data, and then according to the usual viewing habits of the user, a recommendation result of the content is obtained to determine whether to recommend movie content to the user. For example, the user a is scored for viewing, the baseline score of the movie is 7, the "tatanik number" is a good department movie, the item score of the movie is increased by 0.5, but the predicted score of the user a output according to the depth residual error network is generally lower than that of others by 0.3, that is, the user a is a harsh user, so that the score of the user a is subtracted by 0.3, and finally the score of the user a on the movie "tatanik number" is 7.2, and then the movie with the score higher than 6.5 is generally viewed by the user a in practice, for example, the user a only views the movie with the score higher than 6.5, and at this time, the movie tatanik number "can be recommended to the user a according to the result of the score of 7.2.
In order to solve the problem of cold start when a new user is introduced, obtaining a recommendation result of content according to the target prediction scoring data further includes the following operations: and when a new user and a new project are introduced, the baseline scoring data is used as target prediction scoring data to obtain a recommendation result of the content.
Specifically, according to the baseline algorithm, baseline score data can be obtained, that is, the public has a reference score for the film viewing, so that when a new user and a new item are introduced, the baseline score data is used as target prediction score data, and content recommendation is performed according to the target prediction score data. For example, when a new user is introduced, without any information about the past of the user, the similarity between the user and other users cannot be evaluated, and at this time, the baseline score data is used as target prediction score data, and what content is recommended to the new user is determined according to the target prediction score data.
Exemplary device
As shown in fig. 11, an embodiment of the present invention provides a content recommendation apparatus based on a heterogeneous feature deep residual network, including: a source data acquisition unit 401 of content to be recommended, a prediction score data acquisition unit 402, and a content recommendation result acquisition unit 403;
a source data acquiring unit 401 of the content to be recommended, configured to acquire source data of the content to be recommended, and perform data conversion processing on the source data to obtain weighted mixed embedded data;
a prediction score data obtaining unit 402, which obtains prediction score data according to the weighted mixed embedded data and the depth residual error network model;
the content recommendation result obtaining unit 403 obtains a recommendation result of the content according to the prediction score data.
Based on the above embodiments, the present invention further provides an intelligent terminal, and a schematic block diagram thereof may be as shown in fig. 12. The intelligent terminal comprises a processor, a memory, a network interface, a display screen and a temperature sensor which are connected through a system bus. Wherein, the processor of the intelligent terminal is used for providing calculation and control capability. The memory of the intelligent terminal comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the intelligent terminal is used for being connected and communicated with an external terminal through a network. The computer program is executed by a processor to implement a method of content recommendation based on a heterogeneous feature deep residual network. The display screen of the intelligent terminal can be a liquid crystal display screen or an electronic ink display screen, and the temperature sensor of the intelligent terminal is arranged inside the intelligent terminal in advance and used for detecting the operating temperature of internal equipment.
It will be understood by those skilled in the art that the schematic diagram of fig. 12 is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation to the intelligent terminal to which the solution of the present invention is applied, and a specific intelligent terminal may include more or less components than those shown in the figure, or combine some components, or have different arrangements of components.
In one embodiment, an intelligent terminal is provided that includes a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for: acquiring source data of content to be recommended, and performing data conversion processing on the source data to obtain weighted mixed embedded data; obtaining prediction scoring data according to the weighted mixed embedded data and the depth residual error network model; and obtaining a recommendation result of the content according to the prediction scoring data.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
In summary, the present invention discloses a content recommendation method, an intelligent terminal, and a storage medium based on a heterogeneous feature deep residual error network, wherein the method includes: acquiring source data of content to be recommended, and performing data conversion processing on the source data to obtain weighted mixed embedded data; obtaining prediction scoring data according to the weighted mixed embedded data and the depth residual error network model; and obtaining a recommendation result of the content according to the prediction scoring data. According to the embodiment of the invention, the source data of the field content is processed, the processed data is input into the residual error network model to obtain the prediction scoring data, and then the field content is accurately recommended according to the prediction scoring data, so that the calculation method is high in efficiency and low in resource occupancy rate, and meanwhile, the diversified data can avoid cold start when a new user is recommended.
It should be understood that the present invention discloses a content recommendation method based on heterogeneous feature deep residual error network, and it should be understood that the application of the present invention is not limited to the above examples, and it is obvious to those skilled in the art that modifications and changes can be made based on the above description, and all such modifications and changes are intended to fall within the scope of the appended claims.
Claims (10)
1. A content recommendation method based on a heterogeneous feature depth residual error network is characterized by comprising the following steps:
acquiring source data of content to be recommended, and performing data conversion processing on the source data to obtain weighted mixed embedded data;
obtaining prediction scoring data according to the weighted mixed embedded data and the depth residual error network model;
and obtaining a recommendation result of the content according to the prediction scoring data.
2. The content recommendation method based on the heterogeneous feature depth residual error network according to claim 1, wherein the performing data transformation processing on the source data to obtain weighted mixed embedded data comprises:
acquiring collaborative relationship data of a user and a project;
extracting a characteristic value of the source data of the content to be recommended to obtain characteristic value data;
embedding the characteristic value data and the collaborative relationship data of the user and the project based on content to obtain embedded data;
and generating a first triplet loss trainer, and inputting the embedded data into the first triplet loss trainer to obtain weighted mixed embedded data.
3. The content recommendation method based on the heterogeneous feature depth residual error network according to claim 2, wherein the extracting the feature value of the source data of the content to be recommended to obtain the feature value data comprises:
according to the mapping matrix, carrying out linear dimensionality reduction on the source data of the content to be recommended to obtain dimensionality reduction matrix data;
and extracting the characteristic value of the dimension reduction matrix data to obtain characteristic value data.
4. The content recommendation method based on heterogeneous feature depth residual error network according to claim 3, wherein the obtaining of prediction score data according to the weighted hybrid embedded data and the depth residual error network model comprises:
according to the weighted mixed embedded data, extracting heterogeneous characteristic data of the weighted mixed embedded data;
linearly splicing the weighted mixed embedded data and the heterogeneous characteristic data to obtain deep residual error network input data;
and inputting the depth residual error network input data into a depth residual error network model to obtain prediction scoring data.
5. The content recommendation method based on the heterogeneous feature depth residual error network according to claim 4, wherein the generation manner of the depth residual error network model is as follows:
acquiring input sample data and output sample data;
generating a second triplet loss trainer, and generating a network model according to the second triplet loss trainer;
inputting the input sample data into a network model to obtain network model output data;
and when the average absolute error rate of the network model output data and the output sample data is less than a preset value, stopping iteration to obtain a deep residual error network model.
6. The content recommendation method based on the heterogeneous feature depth residual error network according to claim 5, wherein the obtaining of the recommendation result of the content according to the prediction score data comprises:
obtaining a Baseline algorithm result according to a Baseline algorithm;
and obtaining a recommendation result of the content according to the baseline algorithm result and the prediction scoring data.
7. The content recommendation method based on the heterogeneous feature depth residual error network according to claim 6, wherein the obtaining of the recommendation result of the content according to the baseline algorithm result and the prediction score data comprises:
obtaining baseline scoring data, user deviation data and project deviation data according to the baseline algorithm result;
adjusting the prediction scoring data according to the baseline scoring data, the user deviation data and the project deviation data to obtain target prediction scoring data;
and obtaining a recommendation result of the content according to the target prediction scoring data.
8. The method according to claim 7, wherein obtaining the recommendation result of the content according to the target prediction score data further comprises:
and when a new user and a new project are introduced, the baseline scoring data is used as target prediction scoring data to obtain a recommendation result of the content.
9. An intelligent terminal comprising a memory, and one or more programs, wherein the one or more programs are stored in the memory, and wherein the one or more programs being configured to be executed by the one or more processors comprises instructions for performing the method of any of claims 1-8.
10. A non-transitory computer-readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the method of any of claims 1-8.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011180629.0A CN112287222B (en) | 2020-10-29 | 2020-10-29 | Content recommendation method based on heterogeneous characteristic depth residual error network |
| PCT/CN2020/136144 WO2022088417A1 (en) | 2020-10-29 | 2020-12-14 | Content recommendation method based on heterogeneous feature deep residual network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011180629.0A CN112287222B (en) | 2020-10-29 | 2020-10-29 | Content recommendation method based on heterogeneous characteristic depth residual error network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112287222A true CN112287222A (en) | 2021-01-29 |
| CN112287222B CN112287222B (en) | 2023-12-15 |
Family
ID=74373921
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011180629.0A Active CN112287222B (en) | 2020-10-29 | 2020-10-29 | Content recommendation method based on heterogeneous characteristic depth residual error network |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112287222B (en) |
| WO (1) | WO2022088417A1 (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107832804A (en) * | 2017-10-30 | 2018-03-23 | 上海寒武纪信息科技有限公司 | A kind of information processing method and Related product |
| 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 |
| CN111488529A (en) * | 2020-06-28 | 2020-08-04 | 腾讯科技（深圳）有限公司 | Information processing method, information processing apparatus, server, and storage medium |
| CN111506811A (en) * | 2020-03-19 | 2020-08-07 | 上海理工大学 | Click rate prediction method based on deep residual error network |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9886945B1 (en) * | 2011-07-03 | 2018-02-06 | Reality Analytics, Inc. | System and method for taxonomically distinguishing sample data captured from biota sources |
| CN109933678B (en) * | 2019-03-07 | 2021-04-06 | 合肥工业大学 | Artwork recommendation method and device, readable medium and electronic equipment |
| CN110674265B (en) * | 2019-08-06 | 2021-03-02 | 上海孚典智能科技有限公司 | Unstructured information oriented feature discrimination and information recommendation system |
| CN111651613B (en) * | 2020-07-08 | 2021-07-27 | 海南大学 | Knowledge graph embedding-based dynamic recommendation method and system |
-
2020
- 2020-10-29 CN CN202011180629.0A patent/CN112287222B/en active Active
- 2020-12-14 WO PCT/CN2020/136144 patent/WO2022088417A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107832804A (en) * | 2017-10-30 | 2018-03-23 | 上海寒武纪信息科技有限公司 | A kind of information processing method and Related product |
| 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 |
| CN111506811A (en) * | 2020-03-19 | 2020-08-07 | 上海理工大学 | Click rate prediction method based on deep residual error network |
| CN111488529A (en) * | 2020-06-28 | 2020-08-04 | 腾讯科技（深圳）有限公司 | Information processing method, information processing apparatus, server, and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| KAIMING HE等: "Deep Residual Learning for Image Recognition", 2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION(CVPR), pages 1 - 12 * |
| 吴敏: "深度学习框架下充电桩的个性化推荐算法研究", 中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑, pages 037 - 207 * |
| 简恒: "基于 Resnet&FM 模型的电影个性化推荐研究", 中国优秀硕士学位论文全文数据库 哲学与人文科学辑, vol. 2020, pages 087 - 452 * |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2022088417A1 (en) | 2022-05-05 |
| CN112287222B (en) | 2023-12-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ren et al. | Semi-supervised deep embedded clustering | |
| Dieng et al. | Topicrnn: A recurrent neural network with long-range semantic dependency | |
| Liang et al. | A survey of recent advances in transfer learning | |
| Wang et al. | A general exponential framework for dimensionality reduction | |
| Liu et al. | Dual self-attention with co-attention networks for visual question answering | |
| Wen et al. | Preparing lessons: Improve knowledge distillation with better supervision | |
| WO2022105117A1 (en) | Method and device for image quality assessment, computer device, and storage medium | |
| Gao et al. | Self-attention driven adversarial similarity learning network | |
| Zhang et al. | Non-negative tri-factor tensor decomposition with applications | |
| Kong et al. | Probabilistic low-rank multitask learning | |
| Zhou et al. | Self-selective attention using correlation between instances for distant supervision relation extraction | |
| CN114783514A (en) | Method for predicting binding affinity of drug molecules and target protein | |
| Wei et al. | Sequential transformer via an outside-in attention for image captioning | |
| Tong et al. | A deep discriminative and robust nonnegative matrix factorization network method with soft label constraint | |
| Zhu et al. | Age estimation algorithm of facial images based on multi-label sorting | |
| CN113221003B (en) | Mixed filtering recommendation method and system based on dual theory | |
| CN105718898A (en) | Face age estimation method and system based on sparse undirected probabilistic graphical model | |
| CN114861757A (en) | Deep clustering method for images based on dual-correlation reduction | |
| Long et al. | Trainable subspaces for low rank tensor completion: Model and analysis | |
| Khan et al. | Unsupervised domain adaptation using fuzzy rules and stochastic hierarchical convolutional neural networks | |
| Chen et al. | Distributed support vector machine in master–slave mode | |
| Jing et al. | Knowledge-enhanced attentive learning for answer selection in community question answering systems | |
| Wang et al. | Bilateral attention network for semantic segmentation | |
| CN112287222B (en) | Content recommendation method based on heterogeneous characteristic depth residual error network | |
| CN116108363A (en) | Incomplete multi-view multi-label classification method and system based on label guidance |
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 |