CN107203558B - Object recommendation method and device, and recommendation information processing method and device - Google Patents
Object recommendation method and device, and recommendation information processing method and device Download PDFInfo
- Publication number
- CN107203558B CN107203558B CN201610154736.3A CN201610154736A CN107203558B CN 107203558 B CN107203558 B CN 107203558B CN 201610154736 A CN201610154736 A CN 201610154736A CN 107203558 B CN107203558 B CN 107203558B
- Authority
- CN
- China
- Prior art keywords
- score
- target
- original
- cumulative value
- probability cumulative
- 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.)
- Active
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
- G06F16/951—Indexing; Web crawling techniques
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Complex Calculations (AREA)
Abstract
The invention relates to an object recommendation method and device and a recommendation information processing method and device, wherein the object recommendation method comprises the following steps: acquiring an original score of each object in the candidate object set; acquiring the cumulative distribution of target score probability preset in the current object recommendation scene; according to the cumulative distribution of the target score probability, mapping the original score order-preserving regression into a target score; selecting corresponding objects from the candidate object set according to the target scores; recommending the selected object. The object recommendation method and device and the recommendation information processing method and device provided by the invention have the advantage that the recommendation result is accurate.
Description
Technical Field
The invention relates to the technical field of computers, in particular to an object recommendation method and device and a recommendation information processing method and device.
Background
The recommendation object is an important way to deliver information to a user, and the recommended objects such as applications, users or commodities can send recommendation information for recommending the objects to a user terminal, and the user terminal responds to the recommendation information. The present object recommendation method generally adopts a specific scoring algorithm to calculate the score of each object in the candidate object set, and then selects a part of objects from the candidate object set as recommendation results according to the scores.
However, in the current object recommendation method, scoring algorithms adopted in calculating scores are different, the calculated scores of the objects are different, and the distribution conditions of the calculated scores are different, so that the probability of object recommendation under different scoring algorithms is different, which causes inaccurate recommendation results and needs to be improved.
Disclosure of Invention
Therefore, it is necessary to provide an object recommendation method and apparatus, and a recommendation information processing method and apparatus, for solving the problem that the recommendation result is not accurate due to different scoring algorithms used in calculating the score in the current object recommendation method.
A method of object recommendation, the method comprising:
acquiring an original score of each object in the candidate object set;
acquiring the cumulative distribution of target score probability preset in the current object recommendation scene;
according to the cumulative distribution of the target score probability, mapping the original score order-preserving regression into a target score;
selecting corresponding objects from the candidate object set according to the target scores;
recommending the selected object.
An object recommendation apparatus, the apparatus comprising:
the original score acquisition module is used for acquiring the original scores of all the objects in the candidate object set;
the target score probability cumulative distribution acquisition module is used for acquiring preset target score probability cumulative distribution in the current object recommendation scene;
the mapping module is used for carrying out order-preserving regression mapping on the original scores into target scores according to the cumulative distribution of the target score probabilities;
the selecting module is used for selecting corresponding objects from the candidate object set according to the target scores;
and the recommending module is used for recommending the selected object.
A recommendation information processing method, the method comprising:
receiving recommendation information of a recommendation object; the recommendation information is selected according to the target score of the object, and the target score is formed by mapping original scores in an order-preserving regression manner according to the probability cumulative distribution of the target score preset in the current object recommendation scene after scoring the object to obtain the original scores;
sorting the recommendation information according to the target scores of the objects;
and displaying the recommendation information according to the sorting order.
A recommendation information processing apparatus, the apparatus comprising:
the recommendation information receiving module is used for receiving recommendation information of a recommendation object; the recommendation information is selected according to the target score of the object, and the target score is formed by mapping original scores in an order-preserving regression manner according to the probability cumulative distribution of the target score preset in the current object recommendation scene after scoring the object to obtain the original scores;
the sorting module is used for sorting the recommendation information according to the target scores of the objects;
and the recommendation information display module is used for displaying the recommendation information according to the sorting sequence.
According to the object recommendation method and device and the recommendation information processing method and device, the objects are graded to obtain corresponding original score values, and the original score values are subjected to order preserving regression and mapped to the target score values according to the cumulative distribution of the target score values preset in the current object recommendation scene, so that the corresponding objects are recommended according to the target score values. Even if different scoring algorithms are adopted to obtain the original scores, the original scores can be mapped to the target scores meeting the cumulative distribution of the target score probabilities in an order-preserving manner, so that the probability distribution of the target scores used by the recommended objects is accurate, and the recommendation results of the recommended objects are accurate according to the target scores.
Drawings
FIG. 1 is a diagram of an application environment of a recommendation system in one embodiment;
fig. 2 is a schematic diagram of the internal structure of the terminal in one embodiment;
FIG. 3 is a diagram illustrating an internal architecture of a server according to an embodiment;
FIG. 4 is a flowchart illustrating a method for object recommendation in one embodiment;
FIG. 5 is a flowchart illustrating the steps of performing order preserving regression mapping of raw scores to target scores based on the cumulative distribution of target score probabilities in one embodiment;
FIG. 6 is a flowchart illustrating the steps of mapping an original score probability cumulative value to a target score probability cumulative value that matches the original score probability cumulative value based on a target score probability cumulative distribution in one embodiment;
FIG. 7 is a graph illustrating a cumulative distribution of raw score probabilities obtained by the GBDT algorithm in one embodiment;
FIG. 8 is a graphical illustration of a cumulative distribution of probabilities of raw scores obtained by mapping in one embodiment;
FIG. 9 is a flowchart illustrating the steps of generating a cumulative distribution of target score probabilities in one embodiment;
FIG. 10 is a flowchart illustrating the steps of generating a cumulative distribution of target score probabilities in another embodiment;
FIG. 11 is a schematic diagram illustrating a comparison of a pre-calibrated target score probability density curve and a post-calibrated target score probability density curve in one embodiment;
FIG. 12 is a flowchart illustrating a method for processing recommendation information in one embodiment;
FIG. 13 is a block diagram showing the structure of an object recommending apparatus according to an embodiment;
FIG. 14 is a block diagram of the structure of a mapping module in one embodiment;
FIG. 15 is a block diagram showing the construction of an object recommending apparatus according to another embodiment;
FIG. 16 is a block diagram showing the construction of an object recommending apparatus according to still another embodiment;
fig. 17 is a block diagram showing a configuration of a recommended information processing apparatus according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention 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 invention and are not intended to limit the invention.
As shown in fig. 1, in one embodiment, a recommendation system is provided, which includes a terminal 110 and a server 120, wherein the terminal 110 includes a mobile terminal, a vehicle-mounted device, a personal computer, and the like, and the mobile terminal includes at least one of a mobile phone, a tablet computer, a smart watch, or a Personal Digital Assistant (PDA), and the like. The servers 120 may be separate physical servers or may be a cluster of physical servers.
As shown in fig. 2, in one embodiment, a terminal 110 is provided that includes a processor, a non-volatile storage medium, an internal memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor has a calculation function and a function of controlling the operation of the terminal 110, and is configured to execute a recommended information processing method. The non-volatile storage medium includes at least one of a magnetic storage medium, an optical storage medium, and a flash memory type storage medium. The nonvolatile storage medium stores an operating system and also stores a recommended information processing apparatus. The recommendation information processing apparatus is used to implement a recommendation information processing method. The network interface is used to connect to the server 120 through a network. The input device can be a physical key or a touch layer overlapped with the display screen, and the touch layer and the display screen form the touch screen.
As shown in FIG. 3, in one embodiment, a server 120 is provided that includes a processor, a non-volatile storage medium, an internal memory, and a network interface connected by a system bus. Wherein the processor has a computing function and a function of controlling the operation of the server 120, the processor being configured to perform an object recommendation method. The non-volatile storage medium includes at least one of a magnetic storage medium, an optical storage medium, and a flash memory type storage medium. The non-volatile storage medium stores an operating system and also stores an object recommendation device. The object recommending device is used for realizing an object recommending method. The network interface is used to connect to the terminal 110 through a network.
As shown in fig. 4, in an embodiment, an object recommendation method is provided, and this embodiment is illustrated by applying the method to the server 120 in fig. 3. The method specifically comprises the following steps:
The candidate object set is a set formed by a plurality of recommendable objects, the objects in the candidate object set can be represented by corresponding object identifiers, and the object identifiers are unique character strings. The objects in the candidate object set may specifically be users, goods, or applications, and the like, and the goods may be virtual goods or physical goods, and the like.
The server can specifically obtain object attributes of multiple dimensions of each object in the candidate object set, so as to perform scoring through a first scoring algorithm according to the object attributes of the multiple dimensions, and obtain an original score. For example, the server may obtain the commodity attributes of multiple dimensions, such as selling price, bargaining price, click times, click rate, bargaining times, conversion rate, and the like, and score the commodity attributes of each dimension and then perform weighting summation to obtain the original score. The raw score is relative to the target score described below, and is the score before the target score is obtained by mapping.
In one embodiment, the server may specifically extract features of samples in the training set, and train using a machine learning algorithm to obtain a prediction model, thereby extracting features of each object in the candidate object set and inputting the features to the prediction model, thereby predicting an original score of each object in the candidate object set. Wherein the predictive model is a function that maps features of an object to be predicted for a score to a predicted score. The machine learning algorithm used may be, for example, a GBDT (Gradient Boosting Decision Tree) algorithm, a CART (Classification and Regression Trees) algorithm, or a support vector machine algorithm.
And step 404, acquiring a preset target score probability cumulative distribution in the current object recommendation scene.
Specifically, the server stores target score probability cumulative distribution corresponding to various object recommendation scenes in advance, so that after the original score is obtained, the target score probability cumulative distribution in the current object recommendation scene is obtained. The object recommendation scene refers to a scene for recommending a specific type of object, such as a scene for recommending a user, a scene for recommending a commodity, or a scene for recommending an application program.
The target score is an accurate score obtained by learning with a machine learning algorithm different from the machine learning algorithm currently calculating the original score, and the target score probability cumulative distribution is data representing the distribution of the target score probability cumulative values. One target score corresponds to one target score probability cumulative value, and the target score probability cumulative value represents the proportion of all target scores which are less than or equal to the corresponding target scores to all target scores in the target score set. The target scores are discrete, and the probability cumulative distribution can be represented by a discrete functional relationship or can be represented in an enumeration manner.
For example, assuming that the target scores in the target score set are 3, 5, 9 and 10 in ascending order, and the probabilities of the corresponding target scores with respect to the target score set are 0.2, 0.4, 0.2 and 0.2 in order, the cumulative values of the probabilities of the corresponding target scores are 0.2, 0.6, 0.8 and 1 in order.
And step 406, mapping the original scores to target scores by order-preserving regression according to the target score probability cumulative distribution.
Specifically, order preservation means that after the original scores are mapped to the target scores, the corresponding target scores retain the magnitude relation of the original scores. For example, the original scores a and B are mapped to target scores A and B, respectively, and if a < B, then A < B. Regression refers to the probability distribution of the target score obtained by mapping the original score, and the target score probability distribution corresponding to the target score probability cumulative distribution is basically consistent.
In one embodiment, the server may specifically sort the original scores in an ascending order or a descending order, and then sequentially adjust each original score to be a corresponding target score according to the sorting order, so that the target scores meet the target score probability cumulative distribution.
And step 408, selecting corresponding objects from the candidate object set according to the target scores.
Specifically, the server may set an appropriate screening condition according to which the object is selected from the candidate object set, the screening condition being constrained according to the target score. The screening condition may be, for example, selecting an object having a target score larger than a preset value, or selecting an object corresponding to a preset number of target scores having a maximum target score, or selecting an object corresponding to a target score matching a randomly generated score.
At step 410, the selected object is recommended.
Specifically, the server may generate recommendation information recommending the picked object and push the recommendation information to the terminal. The recommendation information comprises object identification, object description text and/or description pictures, and target scores corresponding to each selected object.
According to the object recommendation method, the objects are graded to obtain corresponding original grading values, the original grading values are subjected to order-preserving regression and are mapped to the target grading values according to the cumulative distribution of the target grading probability preset in the current object recommendation scene, and therefore the corresponding objects are recommended according to the target grading values. Even if different scoring algorithms are adopted to obtain the original scores, the original scores can be mapped to the target scores meeting the cumulative distribution of the target score probabilities in an order-preserving manner, so that the probability distribution of the target scores used by the recommended objects is accurate, and the recommendation results of the recommended objects are accurate according to the target scores.
As shown in fig. 5, in an embodiment, the step 406 specifically includes the following steps:
and 502, sequencing the original scores in an ascending order, and obtaining the probability cumulative value of the original score corresponding to each original score according to the sequencing result.
The sorting of the original scores in ascending order means that the original scores are sorted from small to large. Specifically, after sorting the original scores in an ascending order, the server traverses the original scores in the sorting result to obtain a probability value corresponding to the currently traversed original scores, so that all probability values from the first to the currently traversed original scores in the sorting result are added to obtain an original score probability cumulative value corresponding to the currently traversed original scores.
For example, see table one:
| raw scores sorted in ascending order | Raw score probability cumulative value |
| 0.001 | 0.0013 |
| 0.002 | 0.0018 |
| 0.004 | 0.0019 |
| 0.006 | 0.0020 |
| ... | ... |
| 0.32 | 1 |
In the first table, the first column is the original scores sorted in ascending order, the second column is the accumulated original score probability values corresponding to the original scores in the first column, the original scores in each row correspond to the accumulated original score probability values one by one, and the accumulated original score probability value in the second column represents the ratio of the original scores less than or equal to the corresponding original scores in the first column to all the original scores.
Specifically, the server may traverse the original score probability cumulative values according to the target score probability cumulative distribution, find equal or similar target score probability cumulative values for the traversed original score probability cumulative values, and then map the traversed original scores to the found target score probability cumulative values. The fact that the target score probability cumulative value is matched with the original score probability cumulative value means that the target score probability cumulative value and the original score probability cumulative value meet the approaching condition. The server may also assign a matching target score probability accumulation to the traversed original score probability accumulation by interpolation.
Step 506, mapping the original score to a target score corresponding to the target score probability cumulative value.
Specifically, the target score probability cumulative value is generated by sorting the target scores in ascending order, the target score probability cumulative value has a one-to-one correspondence with the target scores, and the server can map the original score to the target score corresponding to the target score probability cumulative value after obtaining the target score probability cumulative value through mapping.
In this embodiment, after the original scores are sorted in an ascending order, the obtained cumulative value of the probability of the original score may reflect the proportion of the original score smaller than or equal to the current original score to all the original scores, and the cumulative value of the probability of the target score may reflect the proportion of the target score smaller than or equal to the corresponding target score to all the target scores. After the original score probability cumulative value and the target score probability cumulative value are matched, the corresponding original score is mapped into the corresponding target score, so that the mapped target score can not only keep the size relation of the original score, but also meet the preset target score probability distribution, and the calculated amount is small.
As shown in fig. 6, in an embodiment, step 504 specifically includes the following steps:
The server may specifically sort, in ascending order, the various target score probability cumulative values represented by the target score probability cumulative distribution, and search for a target score probability cumulative value equal to the original score probability cumulative value in the sorted result.
Specifically, if a target score probability cumulative value equal to an original score probability cumulative value can be found, it is described that the position of a target score corresponding to the found target score probability cumulative value in the entire target score set is the same as the position of an original score corresponding to the corresponding original score probability cumulative value in the entire original score set, and the original score probability cumulative value is directly mapped to the found target score probability cumulative value, so that order preserving regression can be achieved.
For example, see table two:
| raw score | Raw score probability cumulative value | Target score probability cumulative value | Target score |
| 0.001 | 0.0013 | 0.0012 | 0.00001 |
| 0.002 | 0.0018 | 0.0015 | 0.00002 |
| 0.004 | 0.0019 | 0.0015 | 0.00002 |
| 0.006 | 0.0020 | 0.0020 | 0.00004 |
| ... | ... | ... | ... |
| 0.32 | 1 | 1 | 0.05432 |
The target score probability cumulative value of the third column in the second table is obtained by table lookup, specifically, the maximum target score probability cumulative value less than or equal to the corresponding original score probability cumulative value is looked up according to the target score probability cumulative distribution. The target score in the fourth column is in one-to-one correspondence with the target score probability cumulative value in the third column.
For example, in the second table, when the cumulative original score probability value is 0.0020, the cumulative target score probability value equal to 0.0020 can be found, and the corresponding original score 0.006 is mapped to the corresponding target score 0.00004.
Specifically, when the server cannot find the target score probability cumulative value equal to the original score probability cumulative value, the server may find the target score probability cumulative value close to the original score probability cumulative value as the mapping target according to the target score probability cumulative distribution.
In one embodiment, step 606 includes: and mapping the original score probability cumulative value to a target score probability cumulative value closest to the original score probability cumulative value. Specifically, the server may find a target score probability cumulative value having the smallest difference from the original score probability cumulative value, and map the original score probability cumulative value to the target score probability cumulative value.
In one embodiment, step 606 includes: the original score probability cumulative value is mapped to a maximum target score probability cumulative value that is less than the original score probability cumulative value.
For example, in the second table, when the original score probability cumulative value is 0.0018, there is no target score probability cumulative value equal to 0.0018, but the maximum target score probability cumulative value smaller than 0.0018 can be found to be 0.0015, and then the corresponding original score 0.002 is mapped to the corresponding target score 0.00002. For example, if the original score probability cumulative value is 0.0019, there is no target score probability cumulative value equal to 0.0019, but the difference can be made to be 0.0015, which is the maximum target score probability cumulative value smaller than 0.0019, and then the corresponding original score 0.004 is mapped to the corresponding target score 0.00002. It can be seen that the original scores 0.0018 and 0.0019 both map to 0.00002, but this locally equal case does not affect the order of the original scores that remain overall, and is still order preserving regression.
In this embodiment, if the original score is obtained by using the GBDT algorithm, the corresponding probability cumulative distribution curve of the original score is shown in fig. 7, and the probability cumulative distribution curve of the mapped target score is shown in fig. 8. The ordinate of fig. 7 and 8 each represents a probability cumulative value, and the abscissa represents scores sorted in ascending order.
In one embodiment, step 606 includes: the original score probability cumulative value is mapped to a minimum target score probability cumulative value that is greater than the original score probability cumulative value.
In one embodiment, step 606 includes: and acquiring an average value between the maximum target score probability cumulative value smaller than the original score probability cumulative value and the minimum target score probability cumulative value larger than the original score probability cumulative value, and mapping the original score probability cumulative value into the average value. The average here may be an arithmetic average or a weighted average, and the weight of the weighted average may be set as needed.
In this embodiment, according to the cumulative distribution of the target score probabilities, the target score probability cumulative value equal to the original score probability cumulative value is preferentially searched for mapping, and if the target score probability cumulative value is not searched for, the target score probability cumulative value close to the original score probability cumulative value is searched for mapping. The method can achieve more accurate mapping of the order-preserving regression with smaller calculation amount, and is simple, efficient and accurate.
In one embodiment, the target score is a probability value obtained by probability prediction of the prediction sample set according to a prediction model, and the prediction model is obtained by training through a Logistic Regression (LR) algorithm according to a training sample set. In the embodiment, the original score is directly mapped to the probability value, and the object can be directly recommended according to the probability value when being recommended, so that the complicated process of mapping the target score to the probability value is avoided. And the probability distribution obtained by predicting the prediction sample set by adopting the prediction model obtained by training the logistic regression algorithm is closer to the real probability distribution.
As shown in fig. 9, in an embodiment, the object recommendation method further includes a step of generating a cumulative distribution of target score probabilities, and specifically includes the following steps:
Specifically, the server may divide all samples in the current object recommendation scenario into two parts, one part being a full training sample set and the other part being a full prediction sample set. The full set of training samples includes positive training samples and negative training samples.
And 904, obtaining a prediction model by training through a logistic regression algorithm according to the full training sample set.
Specifically, the server may adopt a full training sample set, extract features from training samples in the full training sample set, and train the training samples according to the extracted features and through a logistic regression algorithm to obtain the prediction model. Generally, the total training sample set has a large number of training samples and long training time, but the probability distribution of the score predicted by the prediction model obtained by training is very accurate.
And 906, performing score prediction on the full prediction sample set according to the prediction model to obtain target score probability distribution.
Specifically, the server extracts features from all the prediction samples in the full prediction sample in the same feature extraction mode as the features of the training samples, inputs the features into the trained prediction model, obtains a target score of each prediction sample, and obtains a target score probability distribution for the predicted target score statistical probability.
Specifically, after sorting the target scores in an ascending order, the server may traverse the target scores in the sorting result, and obtain a probability value corresponding to the currently traversed target score, thereby adding all probability values from the top to the currently traversed target score in the sorting result, obtaining a target score probability cumulative value corresponding to the currently traversed target score, and forming a target score probability distribution according to the target score probability cumulative values corresponding to all target scores.
In the embodiment, the prediction model is obtained through training of the logistic regression algorithm according to the full training sample set, so that the value predicted by the trained prediction model is close to the real probability distribution, and the recommendation result is more accurate when an object is recommended.
As shown in fig. 10, in an embodiment, the object recommendation method further includes a step of generating a cumulative distribution of target score probabilities, and specifically includes the following steps:
And 1004, uniformly sampling the full training sample set, and training by a logistic regression algorithm according to the training samples obtained by sampling to obtain a prediction model.
In one embodiment, step 1004 includes: and uniformly sampling the total negative training samples in the total training sample set, and training by a logistic regression algorithm according to the total positive training samples in the total training sample set and the sampled negative training samples to obtain a prediction model.
The quantity of the negative training samples in the full training sample set is far larger than that of the positive training samples, only the full positive training samples are reserved, the prediction accuracy of the trained prediction model can be guaranteed, and the training efficiency can be improved by uniformly sampling the full negative training samples. For example, the total number of training samples in the full training sample set is 1 hundred million, wherein 100 ten thousand positive training samples and 9900 ten thousand negative training samples are provided. If all training samples are used to train the predictive model, the computational power requirements are high. Therefore, the positive training samples are all reserved, and the negative training samples are uniformly sampled according to the proportion of 10%, namely 990 ten thousand training samples are extracted from 9900 ten thousand negative training samples according to a certain rule.
In one embodiment, step 1004 includes: and uniformly sampling the total positive training samples and the total negative training samples in the total training sample set according to different sampling rates, and training through a logistic regression algorithm according to the positive training samples and the negative training samples obtained through sampling to obtain a prediction model. When the training sample set is very large, the full positive training samples and the full negative training samples can be uniformly sampled respectively, so that the training efficiency is improved.
In one embodiment, if the full number of positive training samples is retained, the full number of negative training samples are uniformly sampled at a preset sampling rate r. The predictive model may be calibrated
Is modified into
Wherein xiAs a characteristic of the input, wiFor the weight of the feature, k and c are constant parameters, r is the sampling rate of the negative training samples, and y is the predicted median score.
In one embodiment, the logistic function of the predictive model may be modified to enable calibration of the predictive model. The specific original logic function is p 1/(1+ e)-y) The modified logistic function is p '═ p/(p + (1-p)/r), where y is the predicted intermediate score, and the target score is output by the modified logistic function p'.
And step 1008, performing score prediction on the full prediction sample set according to the calibrated prediction model to obtain target score probability distribution.
In the embodiment, the training efficiency is improved by uniformly sampling the full training sample set, and the target score probability distribution finally obtained by prediction conforms to the real probability distribution by calibrating the prediction model. Referring to fig. 11, curve 1 is the target score probability density curve before calibration, curve 2 is the target score probability density curve after calibration, and curve 2 better conforms to the true target score probability distribution. In fig. 11, the abscissa is the value of the target score, and the ordinate is the corresponding probability value.
As shown in fig. 12, in one embodiment, a recommended information processing method is provided, and this embodiment is exemplified by applying the method to the terminal 110 in fig. 1 and 2 described above. The method specifically comprises the following steps:
Specifically, the server may obtain an original score of each object in the candidate object set; acquiring the cumulative distribution of target score probability preset in the current object recommendation scene; according to the probability cumulative distribution of the target scores, mapping the original scores into the target scores by order-preserving regression; selecting corresponding objects from the candidate object set according to the target scores; and sending recommendation information for recommending the selected object to the terminal. The recommendation information comprises object identification, object description text and/or description pictures, and target scores corresponding to each selected object.
In one embodiment, the server can sort the original scores in ascending order and obtain the probability cumulative value of the original score corresponding to each original score according to the sorting result; mapping the original score probability cumulative value to a target score probability cumulative value matched with the original score probability cumulative value according to the target score probability cumulative distribution; and mapping the original score to a target score corresponding to the target score probability cumulative value.
In one embodiment, the server may find a target score probability cumulative value equal to the original score probability cumulative value according to the target score probability cumulative distribution; if the value is found, mapping the original score probability cumulative value to a found target score probability cumulative value; and if the value is not found, mapping the original score probability cumulative value to a found target score probability cumulative value.
In one embodiment, if the server is not found, the server may map the original score probability cumulative value to a target score probability cumulative value closest to the original score probability cumulative value; or mapping the original score probability cumulative value to a maximum target score probability cumulative value smaller than the original score probability cumulative value; or mapping the original score probability cumulative value to a minimum target score probability cumulative value which is greater than the original score probability cumulative value; or obtaining the average value between the maximum target score probability cumulative value smaller than the original score probability cumulative value and the minimum target score probability cumulative value larger than the original score probability cumulative value, and mapping the original score probability cumulative value into the average value.
In one embodiment, the target score is a probability value obtained by probability prediction of the prediction sample set according to a prediction model, and the prediction model is obtained by training through a logistic regression algorithm according to a training sample set.
In one embodiment, a server may obtain a full training sample set and a full prediction sample set in a current object recommendation scenario; obtaining a prediction model through training of a logistic regression algorithm according to a full training sample set; performing score prediction on the full prediction sample set according to a prediction model to obtain target score probability distribution; and obtaining corresponding target score probability cumulative distribution according to the target score probability distribution.
In one embodiment, a server may obtain a full training sample set and a full prediction sample set in a current object recommendation scenario; uniformly sampling a full training sample set, and training by a logistic regression algorithm according to training samples obtained by sampling to obtain a prediction model; calibrating the prediction model; performing score prediction on the full prediction sample set according to the calibrated prediction model to obtain target score probability distribution; and obtaining corresponding target score probability cumulative distribution according to the target score probability distribution.
And step 1204, sorting the recommendation information according to the target scores of the objects.
And step 1206, displaying the recommendation information according to the sorting order.
Specifically, the terminal may sort the corresponding recommendation information according to the descending order of the target scores, and display the recommendation information according to the descending order of the corresponding target scores according to the sorting order.
The terminal can also obtain an operation instruction for the displayed recommendation information and respond to the recommendation information according to the operation instruction. For example, if the object is a user, the terminal may initiate a friend adding request to the server; if the object is a commodity, the terminal can initiate a commodity purchase request to the server; if the object is an application program, the terminal can initiate an application program downloading request and the like to the server.
According to the recommendation information processing method, the objects are graded to obtain corresponding original grading values, the original grading values are subjected to order-preserving regression and are mapped to the target grading values according to the cumulative distribution of the target grading values preset in the current object recommendation scene, and therefore the corresponding objects are recommended according to the target grading values. Even if different scoring algorithms are adopted to obtain the original scores, the original scores can be mapped to the target scores meeting the cumulative distribution of the target score probabilities in an order-preserving manner, so that the probability distribution of the target scores used by the recommended objects is accurate, and the recommendation results of the recommended objects are accurate according to the target scores.
As shown in fig. 13, in one embodiment, an object recommendation apparatus 1300 is provided, which includes an original score obtaining module 1301, a target score probability cumulative distribution obtaining module 1302, a mapping module 1303, a selecting module 1304, and a recommending module 1305.
An original score obtaining module 1301, configured to obtain an original score of each object in the candidate object set.
A target score probability cumulative distribution obtaining module 1302, configured to obtain a preset target score probability cumulative distribution in a current object recommendation scenario.
And the mapping module 1303 is used for performing order-preserving regression mapping on the original scores to obtain the target scores according to the target score probability cumulative distribution.
And a selecting module 1304, configured to select a corresponding object from the candidate object set according to the target score.
A recommending module 1305, configured to recommend the selected object.
The object recommendation device 1300 scores the object to obtain a corresponding original score value, and then performs order-preserving regression on the original score value according to cumulative distribution of target score values preset in a current object recommendation scene to map the original score value to the target score value, so as to recommend the corresponding object according to the target score value. Even if different scoring algorithms are adopted to obtain the original scores, the original scores can be mapped to the target scores meeting the cumulative distribution of the target score probabilities in an order-preserving manner, so that the probability distribution of the target scores used by the recommended objects is accurate, and the recommendation results of the recommended objects are accurate according to the target scores.
As shown in fig. 14, in one embodiment, the mapping module 1303 includes: an original score probability cumulative value obtaining module 1303a, a probability cumulative value mapping module 1303b and a score mapping module 1303 c.
And the original score probability cumulative value obtaining module 1303a is configured to sort the original scores in an ascending order, and obtain an original score probability cumulative value corresponding to each original score according to a sorting result.
And a probability cumulative value mapping module 1303b, configured to map the original score probability cumulative value to a target score probability cumulative value matched with the original score probability cumulative value according to the target score probability cumulative distribution.
And the score mapping module 1303c is used for mapping the original score to a target score corresponding to the target score probability cumulative value.
In this embodiment, after the original scores are sorted in an ascending order, the obtained cumulative value of the probability of the original score may reflect the proportion of the original score smaller than or equal to the current original score to all the original scores, and the cumulative value of the probability of the target score may reflect the proportion of the target score smaller than or equal to the corresponding target score to all the target scores. After the original score probability cumulative value and the target score probability cumulative value are matched, the corresponding original score is mapped into the corresponding target score, so that the mapped target score can not only keep the size relation of the original score, but also meet the preset target score probability distribution, and the calculated amount is small.
In an embodiment, the probability cumulative value mapping module 1303b is specifically configured to search, according to the target score probability cumulative distribution, a target score probability cumulative value equal to the original score probability cumulative value; if the value is found, mapping the original score probability cumulative value to a found target score probability cumulative value; if the value is not found, mapping the original score probability cumulative value to a target score probability cumulative value closest to the original score probability cumulative value; or mapping the original score probability cumulative value to a maximum target score probability cumulative value smaller than the original score probability cumulative value; or mapping the original score probability cumulative value to a minimum target score probability cumulative value which is greater than the original score probability cumulative value; or obtaining the average value between the maximum target score probability cumulative value smaller than the original score probability cumulative value and the minimum target score probability cumulative value larger than the original score probability cumulative value, and mapping the original score probability cumulative value into the average value.
In this embodiment, according to the cumulative distribution of the target score probabilities, the target score probability cumulative value equal to the original score probability cumulative value is preferentially searched for mapping, and if the target score probability cumulative value is not searched for, the target score probability cumulative value close to the original score probability cumulative value is searched for mapping. The method can achieve more accurate mapping of the order-preserving regression with smaller calculation amount, and is simple, efficient and accurate.
In one embodiment, the target score is a probability value obtained by probability prediction of the prediction sample set according to a prediction model, and the prediction model is obtained by training through a logistic regression algorithm according to a training sample set.
As shown in fig. 15, in one embodiment, the object recommending apparatus 1300 further includes: a full training sample set acquisition module 1306, a training module 1307, a prediction module 1308, and a probability cumulative distribution generation module 1309.
And a full training sample set obtaining module 1306, configured to obtain a full training sample set and a full prediction sample set in a current object recommendation scene.
And a training module 1307, configured to train the prediction model according to the full training sample set and through a logistic regression algorithm.
The predicting module 1308 is configured to perform score prediction on the full prediction sample set according to the prediction model to obtain a target score probability distribution.
A probability cumulative distribution generating module 1309, configured to obtain a corresponding target score probability cumulative distribution according to the target score probability distribution.
In the embodiment, the prediction model is obtained through training of the logistic regression algorithm according to the full training sample set, so that the value predicted by the trained prediction model is close to the real probability distribution, and the recommendation result is more accurate when an object is recommended.
As shown in fig. 16, in one embodiment, the object recommendation apparatus 1300 further includes: a full training sample set acquisition module 1306, a training module 1307, a prediction module 1308, a probability cumulative distribution generation module 1309, and a calibration module 1310.
And a full training sample set obtaining module 1306, configured to obtain a full training sample set and a full prediction sample set in a current object recommendation scene.
And the training module 1307 is configured to perform uniform sampling on the full training sample set, and obtain a prediction model through training of a logistic regression algorithm according to the training samples obtained through sampling.
A calibration module 1310 that calibrates the prediction model.
The predicting module 1308 is configured to perform score prediction on the full prediction sample set according to the calibrated prediction model, so as to obtain a target score probability distribution.
A probability cumulative distribution generating module 1309, configured to obtain a corresponding target score probability cumulative distribution according to the target score probability distribution.
In the embodiment, the training efficiency is improved by uniformly sampling the full training sample set, and the target score probability distribution finally obtained by prediction conforms to the real probability distribution by calibrating the prediction model.
In one embodiment, the training module 1307 is specifically configured to uniformly sample a total number of negative training samples in the total training sample set, and obtain the prediction model by training through a logistic regression algorithm according to the total number of positive training samples in the total training sample set and the sampled negative training samples.
As shown in fig. 17, in one embodiment, there is provided a recommendation information processing apparatus 1700 including: a recommendation information receiving module 1701, a sorting module 1702 and a recommendation information presentation module 1703.
A recommendation information receiving module 1701 for receiving recommendation information of a recommendation object; the recommendation information is selected according to the target score of the object, and the target score is formed by mapping original score order-preserving regression according to the probability cumulative distribution of the target score preset in the current object recommendation scene after scoring the object to obtain the original score.
A sorting module 1702, configured to sort the recommendation information according to the target score of the object.
And a recommendation information presentation module 1703, configured to present recommendation information according to the sorting order.
The recommendation information processing apparatus 1700 scores the object to obtain a corresponding original score, and performs order-preserving regression on the original score to map the original score to the target score according to cumulative distribution of target score probabilities preset in a current object recommendation scene, so as to recommend the corresponding object according to the target score. Even if different scoring algorithms are adopted to obtain the original scores, the original scores can be mapped to the target scores meeting the cumulative distribution of the target score probabilities in an order-preserving manner, so that the probability distribution of the target scores used by the recommended objects is accurate, and the recommendation results of the recommended objects are accurate according to the target scores.
In one embodiment, the target score is a probability value obtained by probability prediction of the prediction sample set according to a prediction model, and the prediction model is obtained by training through a logistic regression algorithm according to a training sample set.
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 a computer program, which can be stored in a computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the computer program is executed. The storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a Random Access Memory (RAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above examples only show some embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (21)
1. A method of object recommendation, the method comprising:
acquiring an original score of each object in the candidate object set; the object comprises at least one of a user, a commodity and an application program; the original scores of the objects are obtained by scoring the object attributes of multiple dimensions of the objects;
acquiring the cumulative distribution of target score probability preset in the current object recommendation scene;
sorting the original scores in an ascending order or a descending order, and sequentially adjusting each original score to be a corresponding target score according to the sorting order so that the target scores meet the target score probability cumulative distribution; the target score is a probability value of occurrence of a predicted event corresponding to the object; when the object is a user, the prediction event is a friend adding event; when the object is a commodity, the predicted event is a commodity purchasing event; when the object is an application program, the predicted event is an application program downloading event;
selecting corresponding objects from the candidate object set according to the target scores;
generating recommendation information corresponding to the selected object, and pushing the recommendation information to a terminal; the recommendation information comprises at least one of an object description text, a description picture and a corresponding target score.
2. The method of claim 1, wherein sorting the raw scores in ascending or descending order and adjusting each raw score in turn in the sorting order to a corresponding target score such that the target score satisfies the target score probability cumulative distribution comprises:
sorting the original scores in ascending order, and obtaining the probability cumulative value of the original score corresponding to each original score according to the sorting result;
mapping the original score probability cumulative value to a target score probability cumulative value matched with the original score probability cumulative value according to the target score probability cumulative distribution;
and mapping the original score to a target score corresponding to the target score probability cumulative value.
3. The method of claim 2, wherein mapping the original score probability accumulation value to a target score probability accumulation value that matches the original score probability accumulation value according to the target score probability accumulation distribution comprises:
searching a target score probability cumulative value equal to the original score probability cumulative value according to the target score probability cumulative distribution;
if the value is found, mapping the original score probability cumulative value to a found target score probability cumulative value;
if not found, then
Mapping the original score probability cumulative value to a target score probability cumulative value closest to the original score probability cumulative value; alternatively, the first and second electrodes may be,
mapping the original score probability cumulative value to a maximum target score probability cumulative value that is less than the original score probability cumulative value; alternatively, the first and second electrodes may be,
mapping the original score probability cumulative value to a minimum target score probability cumulative value that is greater than the original score probability cumulative value; alternatively, the first and second electrodes may be,
and acquiring an average value between a maximum target score probability cumulative value smaller than the original score probability cumulative value and a minimum target score probability cumulative value larger than the original score probability cumulative value, and mapping the original score probability cumulative value to the average value.
4. The method of claim 1, wherein the target score is a probability value obtained by probabilistic prediction of a prediction sample set based on a prediction model, and wherein the prediction model is trained based on a training sample set and by a logistic regression algorithm.
5. The method of claim 1, further comprising:
acquiring a full training sample set and a full prediction sample set under a current object recommendation scene;
obtaining a prediction model through logistic regression algorithm training according to the full training sample set;
performing score prediction on the full prediction sample set according to the prediction model to obtain target score probability distribution;
and obtaining corresponding target score probability cumulative distribution according to the target score probability distribution.
6. The method of claim 1, further comprising:
acquiring a full training sample set and a full prediction sample set under a current object recommendation scene;
uniformly sampling the full training sample set, and training by a logistic regression algorithm according to training samples obtained by sampling to obtain a prediction model;
calibrating the predictive model;
performing score prediction on the full prediction sample set according to the calibrated prediction model to obtain target score probability distribution;
and obtaining corresponding target score probability cumulative distribution according to the target score probability distribution.
7. The method of claim 6, wherein the uniformly sampling the full training sample set, and obtaining the prediction model according to the training samples obtained by sampling and training through a logistic regression algorithm, comprises:
and uniformly sampling the total negative training samples in the total training sample set, and training by a logistic regression algorithm according to the total positive training samples in the total training sample set and the sampled negative training samples to obtain a prediction model.
8. A recommendation information processing method, the method comprising:
receiving recommendation information of a recommendation object; the recommendation information is selected according to target scores of all objects, the target scores are obtained by sorting the original scores in an ascending order or a descending order after the objects are scored to obtain the original scores and sequentially adjusting each original score according to the sorting order, wherein the adjusted corresponding target scores meet the probability cumulative distribution of the target scores preset in the current object recommendation scene; wherein the object comprises at least one of a user, a commodity, and an application; the original scores of the objects are obtained by scoring the object attributes of multiple dimensions of the objects; the target score is a probability value of occurrence of a predicted event corresponding to the object; when the object is a user, the prediction event is a friend adding event; when the object is a commodity, the predicted event is a commodity purchasing event; when the object is an application program, the predicted event is an application program downloading event;
sorting the recommendation information according to the target scores of the objects;
displaying the recommendation information according to a sorting order; the recommendation information includes at least one of an object description text, a description picture, and a target score.
9. The method of claim 8, wherein the target score is a probability value obtained by probabilistic prediction of a prediction sample set based on a prediction model, and wherein the prediction model is trained based on a training sample set and using a logistic regression algorithm.
10. An object recommendation apparatus, characterized in that the apparatus comprises:
the original score acquisition module is used for acquiring the original scores of all the objects in the candidate object set; the object comprises at least one of a user, a commodity and an application program; the original scores of the objects are obtained by scoring the object attributes of multiple dimensions of the objects;
the target score probability cumulative distribution acquisition module is used for acquiring preset target score probability cumulative distribution in the current object recommendation scene;
the mapping module is used for sequencing the original scores in an ascending order or a descending order and sequentially adjusting each original score to be a corresponding target score according to the sequencing order so that the target scores meet the probability cumulative distribution of the target scores; the target score is a probability value of occurrence of a predicted event corresponding to the object; when the object is a user, the prediction event is a friend adding event; when the object is a commodity, the predicted event is a commodity purchasing event; when the object is an application program, the predicted event is an application program downloading event;
the selecting module is used for selecting corresponding objects from the candidate object set according to the target scores;
the recommendation module is used for generating recommendation information corresponding to the selected object and pushing the recommendation information to the terminal; the recommendation information comprises at least one of an object description text, a description picture and a corresponding target score.
11. The apparatus of claim 10, wherein the mapping module comprises:
the original score probability cumulative value acquisition module is used for sequencing the original scores in ascending order and acquiring the original score probability cumulative value corresponding to each original score according to the sequencing result;
a probability cumulative value mapping module, configured to map the original score probability cumulative value to a target score probability cumulative value that matches the original score probability cumulative value according to the target score probability cumulative distribution;
and the score mapping module is used for mapping the original score to a target score corresponding to the target score probability cumulative value.
12. The apparatus according to claim 11, wherein the probability cumulative value mapping module is specifically configured to find a target score probability cumulative value equal to the original score probability cumulative value according to the target score probability cumulative distribution; if the value is found, mapping the original score probability cumulative value to a found target score probability cumulative value; if the value is not found, mapping the original score probability cumulative value to a target score probability cumulative value closest to the original score probability cumulative value; or mapping the original score probability cumulative value to a maximum target score probability cumulative value smaller than the original score probability cumulative value; or mapping the original score probability cumulative value to a minimum target score probability cumulative value greater than the original score probability cumulative value; or obtaining an average value between a maximum target score probability cumulative value smaller than the original score probability cumulative value and a minimum target score probability cumulative value larger than the original score probability cumulative value, and mapping the original score probability cumulative value to the average value.
13. The apparatus of claim 10, wherein the target score is a probability value obtained by probabilistic prediction of a prediction sample set based on a prediction model, and wherein the prediction model is trained based on a training sample set and using a logistic regression algorithm.
14. The apparatus of claim 10, further comprising:
the system comprises a full training sample set acquisition module, a full prediction sample set acquisition module and a full prediction sample set acquisition module, wherein the full training sample set acquisition module is used for acquiring a full training sample set and a full prediction sample set in a current object recommendation scene;
the training module is used for training according to the full training sample set and through a logistic regression algorithm to obtain a prediction model;
the prediction module is used for carrying out score prediction on the full prediction sample set according to the prediction model to obtain target score probability distribution;
and the probability cumulative distribution generating module is used for obtaining corresponding target score probability cumulative distribution according to the target score probability distribution.
15. The apparatus of claim 10, further comprising:
the system comprises a full training sample set acquisition module, a full prediction sample set acquisition module and a full prediction sample set acquisition module, wherein the full training sample set acquisition module is used for acquiring a full training sample set and a full prediction sample set in a current object recommendation scene;
the training module is used for uniformly sampling the full training sample set, and obtaining a prediction model according to the training samples obtained by sampling and through training of a logistic regression algorithm;
a calibration module to calibrate the prediction model;
the prediction module is used for carrying out score prediction on the full prediction sample set according to the calibrated prediction model to obtain target score probability distribution;
and the probability cumulative distribution generating module is used for obtaining corresponding target score probability cumulative distribution according to the target score probability distribution.
16. The apparatus of claim 15, wherein the training module is specifically configured to uniformly sample a total number of negative training samples in the total number of training sample sets, and obtain the prediction model according to the total number of positive training samples in the total number of training sample sets and the sampled negative training samples and through training with a logistic regression algorithm.
17. A recommended information processing apparatus, characterized in that the apparatus comprises:
the recommendation information receiving module is used for receiving recommendation information of a recommendation object; the recommendation information is selected according to target scores of all objects, the target scores are obtained by sorting the original scores in an ascending order or a descending order after the objects are scored to obtain the original scores and sequentially adjusting each original score according to the sorting order, wherein the adjusted corresponding target scores meet the probability cumulative distribution of the target scores preset in the current object recommendation scene; wherein the object comprises at least one of a user, a commodity, and an application; the original scores of the objects are obtained by scoring the object attributes of multiple dimensions of the objects; the target score is a probability value of occurrence of a predicted event corresponding to the object; when the object is a user, the prediction event is a friend adding event; when the object is a commodity, the predicted event is a commodity purchasing event; when the object is an application program, the predicted event is an application program downloading event;
the sorting module is used for sorting the recommendation information according to the target scores of the objects;
the recommendation information display module is used for displaying the recommendation information according to the sorting sequence; the recommendation information includes at least one of an object description text, a description picture, and a target score.
18. The apparatus of claim 17, wherein the target score is a probability value obtained by probabilistic prediction of a prediction sample set based on a prediction model, and wherein the prediction model is trained based on a training sample set and using a logistic regression algorithm.
19. A server comprising a memory and a processor, the memory storing a computer program, characterized in that the processor realizes the steps of the method of any one of claims 1 to 7 when executing the computer program.
20. A terminal comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 8 to 9.
21. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610154736.3A CN107203558B (en) | 2016-03-17 | 2016-03-17 | Object recommendation method and device, and recommendation information processing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610154736.3A CN107203558B (en) | 2016-03-17 | 2016-03-17 | Object recommendation method and device, and recommendation information processing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107203558A CN107203558A (en) | 2017-09-26 |
| CN107203558B true CN107203558B (en) | 2021-03-09 |
Family
ID=59904136
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610154736.3A Active CN107203558B (en) | 2016-03-17 | 2016-03-17 | Object recommendation method and device, and recommendation information processing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107203558B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816187B (en) * | 2017-11-21 | 2022-09-23 | 财付通支付科技有限公司 | Information processing method, information processing device, computer equipment and storage medium |
| CN108898591A (en) * | 2018-06-22 | 2018-11-27 | 北京小米移动软件有限公司 | Methods of marking and device, electronic equipment, the readable storage medium storing program for executing of picture quality |
| CN111382349B (en) * | 2018-12-29 | 2023-08-15 | 广州市百果园网络科技有限公司 | Information recommendation method, device, computer equipment and storage medium |
| CN112925976B (en) * | 2021-01-29 | 2023-11-21 | 北京达佳互联信息技术有限公司 | Method, device, server and storage medium for allocating denomination data |
| CN113256743B (en) * | 2021-06-16 | 2022-09-02 | 图兮数字科技(北京)有限公司 | Image processing method and device, electronic equipment and readable storage medium |
| CN114528482B (en) * | 2022-01-25 | 2022-12-27 | 北京三快在线科技有限公司 | Method and device for determining recommended object, electronic equipment and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102594905A (en) * | 2012-03-07 | 2012-07-18 | 南京邮电大学 | Method for recommending social network position interest points based on scene |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8942917B2 (en) * | 2011-02-14 | 2015-01-27 | Microsoft Corporation | Change invariant scene recognition by an agent |
| CN102508870B (en) * | 2011-10-10 | 2013-09-11 | 南京大学 | Individualized recommending method in combination of rating data and label data |
| US8926436B2 (en) * | 2012-06-22 | 2015-01-06 | Disney Enterprises, Inc. | Method and device for fantasy sports roster recommendations |
| CN103927347A (en) * | 2014-04-01 | 2014-07-16 | 复旦大学 | Collaborative filtering recommendation algorithm based on user behavior models and ant colony clustering |
-
2016
- 2016-03-17 CN CN201610154736.3A patent/CN107203558B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102594905A (en) * | 2012-03-07 | 2012-07-18 | 南京邮电大学 | Method for recommending social network position interest points based on scene |
Non-Patent Citations (1)
| Title |
|---|
| 基于位置的社会化网络推荐系统;刘树栋 等;《计算机学报》;20150228;第38卷(第2期);322-336 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107203558A (en) | 2017-09-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107203558B (en) | Object recommendation method and device, and recommendation information processing method and device | |
| CN106651542B (en) | Article recommendation method and device | |
| CN110866181B (en) | Resource recommendation method, device and storage medium | |
| CN106355449B (en) | User selection method and device | |
| CN110909182B (en) | Multimedia resource searching method, device, computer equipment and storage medium | |
| CN110232403B (en) | Label prediction method and device, electronic equipment and medium | |
| CN107357793B (en) | Information recommendation method and device | |
| CN109360057B (en) | Information pushing method, device, computer equipment and storage medium | |
| CN112328909B (en) | Information recommendation method and device, computer equipment and medium | |
| WO2020192013A1 (en) | Directional advertisement delivery method and apparatus, and device and storage medium | |
| CN108304354B (en) | Prediction model training method and device, storage medium and electronic equipment | |
| CN112395487B (en) | Information recommendation method and device, computer readable storage medium and electronic equipment | |
| CN110647696A (en) | Business object sorting method and device | |
| CN111178949A (en) | Service resource matching reference data determination method, device, equipment and storage medium | |
| WO2024041483A1 (en) | Recommendation method and related device | |
| CN115062732A (en) | Resource sharing cooperation recommendation method and system based on big data user tag information | |
| Ben-Shimon et al. | An ensemble method for top-N recommendations from the SVD | |
| CN111429161B (en) | Feature extraction method, feature extraction device, storage medium and electronic equipment | |
| CN109656433B (en) | Category information processing method, category information processing device, computer equipment and storage medium | |
| Meng et al. | Merged region based image retrieval | |
| CN110909193A (en) | Image sorting display method, system, equipment and storage medium | |
| CN112783468A (en) | Target object sorting method and device | |
| CN114756758B (en) | Hybrid recommendation method and system | |
| CN113868543B (en) | Method for sorting recommended objects, method and device for model training and electronic equipment | |
| CN112905885B (en) | Method, apparatus, device, medium and program product for recommending resources to user |
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 |