CN109063104B - Recommendation information refreshing method and device, storage medium and terminal equipment - Google Patents
Recommendation information refreshing method and device, storage medium and terminal equipment Download PDFInfo
- Publication number
- CN109063104B CN109063104B CN201810846642.1A CN201810846642A CN109063104B CN 109063104 B CN109063104 B CN 109063104B CN 201810846642 A CN201810846642 A CN 201810846642A CN 109063104 B CN109063104 B CN 109063104B
- Authority
- CN
- China
- Prior art keywords
- information
- sequence
- recommendation
- recommendation information
- refreshing
- 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
Abstract
The invention provides a method and a device for refreshing recommendation information, a storage medium and terminal equipment, wherein the method comprises the following steps: responding to a refreshing request of a request terminal, extracting recommendation information from the recommendation information set to combine into an initial information sequence; wherein the initial information sequence comprises M pieces of recommendation information; adjusting the arrangement sequence of the initial information sequence according to the predicted click probability of each piece of recommended information in the initial information sequence to obtain an optimal information sequence; extracting a recommendation information sequence from the preference information sequence, wherein the recommendation information sequence comprises top N pieces of recommendation information in the preference information sequence, and M is greater than N; and sending the recommendation information sequence to the request end to refresh and display at the request end. By adopting the invention, the recommendation information sequence issued in the future refreshing can be considered on the premise that the recommendation information sequence issued in the refreshing is optimal.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a method and a device for refreshing recommendation information, a storage medium and terminal equipment.
Background
With the development of the internet, various recommendation platforms emerge, and information is pushed or recommended to internet users. For example, the product offer information is pushed, search information is recommended by a search engine, and information such as news information and articles in various fields is pushed.
The recommendation system of the recommendation platform issues recommendation information by taking refreshing as a unit. For example, when a user triggers a refresh operation at a client, a request is sent to a server. The server side responds to the request and forms the recommendation list with the n pieces of recommendation information to be issued to the client side. The client displays the received recommendation list. The server side enables the n pieces of recommendation information and the arrangement sequence issued to the user to match the interests of the user as much as possible through a recommendation algorithm so as to attract the user to click and read. However, the recommendation system usually only considers the information sequence in the current refresh as optimal as possible, and the optimization of the information sequence between adjacent refreshes is rarely considered. For example, the user requests 3 refreshes, resulting in list (1), list (2), and list (3), respectively. If only the recommendation information in the current refresh is referred to, the recommendation information may be the optimal list, i.e. the local optimal list, of the current refresh. However, if 3 lists are spliced together into a composed list, the composed list is not necessarily a global optimal list. Particularly, when the user performs pull-up refreshing, the issued lists which can be refreshed for a plurality of times in history are spliced together for displaying. At this time, the list displayed after being concatenated is not necessarily the most optimally arranged list. Therefore, how to make the list composed of the issued lists refreshed for many times also be the list with the optimal arrangement is a technical problem which needs to be solved at present.
The conventional technical scheme is as follows: and selecting and arranging the issued recommendation information in the current refreshing through a recommendation algorithm based on session. In the process, recommendation information issued by multiple refreshes in the past and the arrangement sequence among the information are taken as characteristics and added into a recommendation algorithm for calculation. Therefore, the obtained calculation result can enable the list displayed by combining the recommendation information list issued by the refreshing and the recommendation information list historically displayed by the client to be optimally arranged.
Although, in the above technical scheme, when issuing the updated recommended information list, only the characteristics of the historically issued recommended information list are considered, the updated list is spliced with the updated list in the future, and is not necessarily the optimal list. In the continuous refreshing process of the user, the browsing experience of the user is influenced.
Disclosure of Invention
Embodiments of the present invention provide a method, an apparatus, a storage medium, and a terminal device for refreshing recommendation information, so as to solve or alleviate one or more of the above technical problems in the prior art.
In a first aspect, an embodiment of the present invention provides a method for refreshing recommendation information, including:
responding to a refreshing request of a request terminal, extracting recommendation information from the recommendation information set to combine into an initial information sequence; wherein the initial information sequence comprises M pieces of recommendation information;
adjusting the arrangement sequence of the initial information sequence according to the predicted click probability of each piece of recommended information in the initial information sequence to obtain an optimal information sequence;
extracting a recommendation information sequence from the preference information sequence, wherein the recommendation information sequence comprises top N pieces of recommendation information in the preference information sequence, and M is greater than N; and
and sending the recommendation information sequence to the request end to refresh and display at the request end.
With reference to the first aspect, in a first implementation manner of the first aspect, the extracting, from the recommendation information set, recommendation information to combine into an initial information sequence in response to a refresh request of a request end includes:
searching corresponding recommendation information from a recommendation information base according to the user information in the refreshing request to obtain a recommendation information set;
calculating recommendation weight values of all recommendation information in the recommendation information set according to a preset recommendation algorithm; and
sorting all recommendation information in the recommendation information set according to recommendation weight values; and
and selecting the top M pieces of recommended information to combine into an initial information sequence.
With reference to the first aspect, in a second implementation manner of the first aspect, the adjusting an arrangement order of the initial information sequence according to a predicted click probability of each piece of recommended information in the initial information sequence to obtain a preferred information sequence includes:
randomly adjusting the arrangement sequence of the recommended information in the initial information sequence to obtain a candidate information sequence;
calculating the predicted click probability of each piece of recommended information of the candidate information sequence through a sequence evaluation model;
calculating recommendation scores of the candidate information sequences according to the predicted click probability of each piece of recommendation information of the candidate information sequences; and
and selecting the candidate information sequence with the highest recommendation score from the obtained candidate information sequences as a preferred information sequence.
With reference to the second implementation manner of the first aspect, in a third implementation manner of the first aspect, the randomly adjusting an arrangement order of the recommendation information in the initial information sequence to obtain a candidate information sequence includes:
performing variation operation on the arrangement positions of the recommended information in the initial information sequence to obtain a candidate information sequence; wherein the mutation operation comprises randomly selecting two pieces of recommended information from the initial information sequence for position exchange.
With reference to the second implementation manner of the first aspect, in a fourth implementation manner of the first aspect, the refresh method further includes:
acquiring training data of the sequence evaluation model; the training data comprises a sample recommendation information sequence of the request terminal and the click probability of each recommendation information in the sample recommendation information sequence; and
and inputting the training data into the sequence evaluation model according to the generation time of the sample recommendation information sequence so as to train and update the sequence evaluation model.
In a second aspect, an embodiment of the present invention further provides a device for refreshing recommendation information, including:
the initial sequence generation module is used for responding to a refreshing request of a request terminal and extracting recommendation information from the recommendation information set to combine the recommendation information into an initial information sequence; wherein the initial information sequence comprises M pieces of recommendation information;
the optimal sequence generation module is used for adjusting the arrangement sequence of the initial information sequence according to the predicted click probability of each piece of recommended information in the initial information sequence to obtain an optimal information sequence;
a recommended sequence extraction module, configured to extract a recommended information sequence from the preferred information sequence, where the recommended information sequence includes top N pieces of recommended information in the preferred information sequence, where M is greater than N; and
and the sequence sending module is used for sending the recommendation information sequence to the request end so as to refresh and display at the request end.
With reference to the second aspect, in a first implementation manner of the second aspect, the initial sequence generation module includes:
the recommendation information searching unit is used for searching corresponding recommendation information from a recommendation information base according to the user information in the refreshing request to obtain a recommendation information set;
the information weight calculation unit is used for calculating recommendation weight values of all recommendation information in the recommendation information set according to a preset recommendation algorithm; and
the recommendation information sorting unit is used for sorting the recommendation information in the recommendation information set according to a recommendation weight value; and
and the recommendation information selecting unit is used for selecting the recommendation information which is sequenced at the top M and combining the recommendation information into an initial information sequence.
With reference to the second aspect, in a second implementation manner of the second aspect, the preferred sequence generation module includes:
a candidate sequence generating unit, configured to randomly adjust an arrangement order of the recommendation information in the initial information sequence to obtain a candidate information sequence;
the click probability calculation unit is used for calculating the predicted click probability of each piece of recommended information of the candidate information sequence through a sequence evaluation model;
the recommendation score calculating unit is used for calculating recommendation scores of the candidate information sequences according to the predicted click probability of each piece of recommendation information of the candidate information sequences; and
and the sequence selection unit is used for selecting the candidate information sequence with the highest recommendation score from the obtained candidate information sequences as the preferred information sequence.
With reference to the second implementation manner of the second aspect, in a third implementation manner of the second aspect, the candidate sequence generating unit is specifically configured to: performing variation operation on the arrangement positions of the recommended information in the initial information sequence to obtain a candidate information sequence; wherein the mutation operation comprises randomly selecting two pieces of recommended information from the initial information sequence for position exchange.
With reference to the second implementation manner of the second aspect, in a fourth implementation manner of the second aspect, the refresh apparatus further includes:
the training data module is used for acquiring training data of the sequence evaluation model; the training data comprises a sample recommendation information sequence of the request terminal and the click probability of each recommendation information in the sample recommendation information sequence; and
and the model training module is used for inputting the training data into the sequence evaluation model according to the generation time of the sample recommendation information sequence so as to train and update the sequence evaluation model.
The functions of the device can be realized by hardware, and can also be realized by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above-described functions.
In one possible design, the recommendation information refreshing structure includes a processor and a memory, the memory is used for the recommendation information refreshing apparatus to execute the recommendation information refreshing program in the first aspect, and the processor is configured to execute the program stored in the memory. The device for refreshing the recommendation information may further include a communication interface, and the device for refreshing the recommendation information may communicate with other devices or a communication network.
In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium for computer software instructions used by a device for refreshing recommended information, where the computer software instructions include a program for executing the method for refreshing recommended information in the first aspect.
One of the above technical solutions has the following advantages or beneficial effects:
when the refresh request is responded, an initial information sequence which is longer than a sent recommendation information sequence is generated, the sequence arrangement order is adjusted to obtain an optimal information sequence by combining the predicted click probability in each recommendation information in the initial information sequence, the top N recommendation information in the sequence is extracted from the optimal information sequence to form the recommendation information sequence, and the recommendation information sequence is sent to a request end to be displayed at the request end. Therefore, on the premise that the recommendation information sequence issued by the refreshing at this time is optimal, the recommendation information sequence issued by the refreshing in the future can be considered. Therefore, in the continuous refreshing process of the user, the displayed content is refreshed to attract the user, and the user obtains better browsing experience.
The foregoing summary is provided for the purpose of description only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the present invention will be readily apparent by reference to the drawings and following detailed description.
Drawings
In the drawings, like reference numerals refer to the same or similar parts or elements throughout the several views unless otherwise specified. The figures are not necessarily to scale. It is appreciated that these drawings depict only some embodiments in accordance with the disclosure and are therefore not to be considered limiting of its scope.
FIG. 1 is a flow chart illustrating an embodiment of a method for refreshing recommendation information provided by the present invention;
FIG. 2 is a flow chart diagram illustrating one embodiment of a process for generating an initial information sequence provided by the present invention;
FIG. 3 is a flow chart diagram illustrating one embodiment of a preferred information sequence acquisition process provided by the present invention;
FIG. 4 is a schematic diagram of one embodiment of a sequence estimation model provided by the present invention;
FIG. 5 is a flowchart of an exemplary application of a refreshing method for a recommended article provided in the present invention;
FIG. 6 is a schematic diagram of one embodiment of the differences between the refreshing method of the recommended article provided by the present invention and the refreshing method of the prior art;
FIG. 7 is a schematic structural diagram of an embodiment of a device for refreshing recommended information provided by the present invention;
FIG. 8 is a schematic structural diagram of an initial sequence generating module of a device for refreshing recommended information according to an embodiment of the present invention;
FIG. 9 is a schematic structural diagram of an embodiment of a preferred sequence generation module of the device for refreshing recommendation information provided by the present invention;
FIG. 10 is a schematic structural diagram of another embodiment of a device for refreshing recommended information provided by the present invention;
fig. 11 is a schematic structural diagram of an embodiment of a terminal device provided by the present invention.
Detailed Description
In the following, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. Accordingly, the drawings and description are to be regarded as illustrative in nature, and not as restrictive.
Referring to fig. 1, an embodiment of the present invention provides a method for refreshing recommendation information, which can be applied to a server device. The server device may include a computer, a microcomputer, or the like. The embodiment includes steps S100 to S400, which are specifically as follows:
s100, responding to a refreshing request of a request end, extracting recommendation information from a recommendation information set and combining the recommendation information into an initial information sequence; wherein the initial information sequence comprises M pieces of recommendation information.
In this embodiment, the requesting end may include a smart phone, a tablet, a computer, and the like. The requesting end may install a client application program matched with the server system for sending a request to the server device and displaying the received information. The client applications may include APPs (applications) such as hundredths, google, nam, and jingdong. The recommendation information may include commodity preference information, search information of a search engine, news information, articles of various fields, and the like. If the refresh request is a request of a user to refresh a retrieval page of a product, the recommendation information may include description information of the product, such as a picture of the product, a price, a manufacturer, and the like. If the refresh request is a request from a user to refresh a search page of a search sentence, the recommendation information may include articles or news information related to the search sentence. The initial information sequence may be arranged according to the degree of correlation between the recommendation information and the user information of the requesting end or the predicted click rate of the recommendation information. The length of the initial information sequence is the number of recommendation information contained in the initial information sequence. The length of the initial information sequence is longer than the length of the recommended information sequence sent to the requesting end.
And S200, adjusting the arrangement sequence of the initial information sequence according to the predicted click probability of each piece of recommended information in the initial information sequence to obtain an optimal information sequence.
In this embodiment, since the arrangement order of the initial information sequence only considers the weight values of the recommendation information in the recommendation information set, which is often not an optimal sequence, the influence of the whole sequence on the user needs to be considered, for example, the click probability of the user on each recommendation information needs to be predicted.
S300, extracting a recommendation information sequence from the preferred information sequence, wherein the recommendation information sequence comprises top N pieces of recommendation information sequenced in the preferred information sequence, and M is larger than N.
In this embodiment, the value of M may be an integer multiple of N. For example, taking the recommended information as an article as an example, if N articles are sent for any refresh request, the preferred information sequence or the initial information sequence may include M ═ N × S articles, where S is greater than 1 and is the number of refreshes that needs to be considered in the future including the current refresh. If N is the value 5 and M is the value 20, the recommendation information sequence is the recommendation information ranked in the top 5 of the preference information sequence.
S400, sending a recommendation information sequence to the request end to refresh and display at the request end.
In this embodiment, in the process of issuing the recommended information sequence in this refresh, a longer initial sequence (initial information sequence) is generated in advance. Then, the arrangement order in the initial sequence is adjusted to obtain a preferred sequence (preferred information sequence). And finally, extracting the top N pieces of recommendation information from the preferred sequence to serve as a distributed recommendation information sequence. Since the preferred sequence length is longer than the delivered sequence, the delivered sequence of this refresh can take into account the delivered sequence of the subsequent refresh.
In a possible implementation manner, as shown in fig. 2, the generating process of the initial information sequence of step S100 may include:
and S110, searching corresponding recommendation information from the recommendation information base according to the user information in the refreshing request to obtain a recommendation information set.
In this embodiment, the refresh request sent by the request end usually carries user information. In order to make the recommendation information more matched with the requirements of the user, the information can be extracted from the recommendation information base by combining the user information while the refreshing request is met. If the same user refreshes continuously for a plurality of times in a short time, the recommendation information which is refreshed by the user in a certain time in the past can be eliminated when the recommendation information is refreshed and searched.
And S120, calculating recommendation weight values of the recommendation information in the recommendation information set according to a preset recommendation algorithm.
In the present embodiment, the recommendation algorithm may include a matrix decomposition-based method, a neural network-based method, and the like. The click probability of each piece of recommendation information in the set or the correlation degree of the recommendation information and the user information can be calculated by adopting a recommendation algorithm. The recommendation weight value may depend on the click probability of the recommendation information, the degree of correlation with the user information, and the like. For example, the higher the click probability of the recommendation information, the higher the recommendation weight value; the higher the degree of correlation of the user information, the higher the recommendation weight value.
And S130, sorting the recommendation information in the recommendation information set according to the recommendation weight value.
S140, selecting the top M pieces of recommendation information in sequence, and combining the recommendation information into an initial information sequence.
For example, if the length of the recommendation information sequence is N, the number of times of refreshing set by the system of the server is S, and S is greater than 1, the top N × S pieces of recommendation information in the recommendation information set may be selected. At this time, M is N × S.
The ranking of the initial information sequence is only based on the recommendation weight value, and the recommendation weight value of each recommendation information is calculated based on the recommendation information, and the relation among the recommendation information cannot be considered, namely the influence of the sequence formed by the recommendation information on the user cannot be considered. Therefore, if the initial information sequence is issued to the user, it is not necessarily the optimal sequence for this refresh, and the sequence is too long. For this reason, the embodiment needs to obtain the preferred information sequence based on the initial information sequence, and intercept a sequence with a suitable length from the preferred information sequence as a sequence issued to the request end.
In a possible implementation manner, as shown in fig. 3, the acquiring process of the preferred information sequence of step S200 may include:
s210, randomly adjusting the arrangement sequence of the recommended information in the initial information sequence to obtain a candidate information sequence.
In this embodiment, the sequence obtained by adjusting the arrangement order of the recommendation information in the initial information sequence is similar to but different from the initial sequence, so that the preferred sequence can be conveniently found from the initial sequence. Specifically, the mutation operation may be performed on the arrangement positions of the recommendation information in the initial information sequence, and the mutation operation includes randomly selecting two recommendation information from the initial information sequence for position exchange. And performing one or more mutation operations to obtain a new sequence different from the original sequence. The above process is executed for multiple times, and the same sequence is excluded, so that multiple sequences can be obtained to form a candidate information sequence set.
Illustratively, if the initial information sequence is { information A1, information A2, information A3, information A4, information A5} arranged in order, the information is randomly selected for location exchange. For example, information A1 is location swapped with information A3. One or more times, for example, two times (the first time the information a1 is location exchanged with the information A3, the second time the information A3 is location exchanged with the information a 5) may be performed randomly. By way of example, the candidate information sequences can be obtained as { information a5, information a2, information a1, information a4, information A3} arranged in order.
S220, calculating the predicted click probability of each piece of recommended information of the candidate information sequence through the sequence evaluation model.
In this embodiment, the sequence estimation model may be constructed based on a neural network, for example: RNN (Recurrent Neural Network), a variation of RNN Network, and the like. As shown in fig. 4, the sequence estimation model can be modeled by means of RNN. In the drawingsThe neural nodes are sequentially expanded from left to right in time sequence, x is an input layer, h is a hidden layer, and y is an output layer. h ist-n，……,，ht-2，ht-1，htNeural nodes, x, representing hidden statest-n,……,xt-2,xt-1,xtFeatures representing recommended information in the sequence, yt-n,……,yt-2,yt-1,ytIndicating the probability of whether each piece of recommendation information in the sequence will click.
In this embodiment, the sequence evaluation model may be trained in advance. For example, BPTT (Back Propagation delay Time) may be used for model training. In general, the training process of the model may include: acquiring training data of a sequence evaluation model; the training data comprises a sample recommendation information sequence of the request terminal and the click probability of each recommendation information in the sample recommendation information sequence. For example, if the recommendation information in the sample is clicked in the sample display process, the click probability of the recommendation information is 1; and if the recommendation information in the sample is not clicked in the sample screen display process, the click probability of the recommendation information is 0. And then, sequentially inputting training data into the sequence evaluation model according to the generation time of the sample recommendation information sequence so as to train and update the sequence evaluation model. It may be noted that the hidden layer of RNNs includes self-connection weights W between hidden state neural networks. Therefore, when the features of the recommendation information are sequentially put into the RNN network for learning, the RNN can learn the relationship between adjacent recommendation information through update iteration of W, thereby implementing modeling of the sequence.
And S230, calculating recommendation scores of the candidate information sequences according to the predicted click probability of each piece of recommendation information of the candidate information sequences.
In this embodiment, the predicted click probabilities of the pieces of recommendation information in the sequence may be added to obtain the recommendation score of the sequence. Each recommendation information in the sequence may also be given a weight value. For example, the weight values corresponding to the recommendation information are given according to the arrangement sequence of the recommendation information, the predicted click probability of each recommendation information is multiplied by the weight value of the recommendation information, and the products are summed to obtain the recommendation score of the sequence.
S240, selecting the candidate information sequence with the highest recommendation score from the obtained candidate information sequences as a preferred information sequence.
In this embodiment, for an arbitrarily input recommendation information sequence, a click probability is predicted for each recommendation information in the sequence by using a sequence evaluation model. Because the RNN learns the relationship between the recommendation information in the sequence and then calculates a recommendation score for the sequence by utilizing the click probability obtained by calculating the sequence evaluation model for each recommendation information, the recommendation score can consider the relationship between the recommendation information in the sequence. Furthermore, the sequence selected from the candidate information sequences according to the recommendation score tends to be better than the initial information sequence.
Referring to fig. 5 and fig. 6, fig. 5 is a flowchart of an application example of a refreshing method for a recommended article according to an embodiment of the present invention. Fig. 6 is a schematic diagram illustrating a difference between a refreshing method of a recommended article provided by an embodiment of the present invention and a refreshing method of the prior art.
The basic principle of the application example is as follows: the server side can recall the article set from the article library every time of refreshing so as to extract the articles from the article set and form an article sequence (list) for issuing. It should be noted that there is also a possibility of recall in subsequent refreshes, and the articles left in the post-chapter set of the article issued in the current refresh are planed. Especially in the case of continuous refresh over a short period of time. Because the user portrait does not change much in a short time, the resources in the article library do not change much, and the possibility that the remaining articles in the article set are recalled is high. Therefore, when the issuing sequence is generated in the refreshing, if the article which is not issued in the refreshing is also taken into consideration, the refreshing issuing sequence and the subsequent refreshing issuing sequence can be spliced together to form an integral optimal sequence. Among them, one possible solution is: in the process of refreshing and issuing the article sequence, a longer sequence is generated, and the refreshing only issues the first N articles in the sequence. Therefore, the article sequence issued by the refreshing at this time takes the article sequence issued by the refreshing in the future into consideration, so that the sequence is the integral optimal sequence when the sequence issued by the refreshing at this time is spliced with the sequence issued by the refreshing in the future.
Referring to fig. 5, the present application example includes 4 steps, specifically as follows:
(1) an initial sequence is generated from the set of recalled articles.
The main purpose of this step is to generate a better initial sequence from the set of articles recalled.
The specific method comprises the following steps:
the method comprises the following steps of calculating a weight w for each article recalled from an article library by adopting any recommendation algorithm, wherein the weight is used for expressing the degree of correlation with user information of a request end or the probability of clicking the article by a user.
And secondly, sorting the articles from large to small according to the weight w of each article, and selecting the articles sorted at the top m × n. Wherein n represents the number of the issued articles refreshed at one time; m >1, indicating the total number of refreshes that need to be considered.
For the recommendation algorithm to calculate the weight w, various classical recommendation algorithms may be employed. Such as matrix factorization based methods, neural network based methods, and the like.
(2) Randomly generating a plurality of sequence combinations from an initial sequence
In this embodiment, the initial sequence tends to be better, but not optimal. Because the rank order in the initial sequence only considers the weight of the article, the calculation of the weight of the article is only based on the information of the article itself, and the relationship with other articles is not considered. That is, the initial sequences formed fail to take into account the article between sequences. The main purpose of this step is to generate a set of sequences from the original sequence that are similar to the original sequence but different from the original sequence in order to find a better sequence.
The specific method comprises the following steps:
for a sequence, this embodiment defines a mutation operation: randomly picking the positions i and j (i ≠ j) in the two sequences, and exchanging the articles of the ith position and the jth position. For the initial sequence, x (x is a random number) mutation operations can be performed to obtain a new sequence. The present embodiment performs the above process multiple times to obtain multiple different candidate sequences.
(3) Selecting an optimal sequence according to an evaluation model
The main purpose of this step is to pick the optimal sequence from a number of different candidate sequences. And scoring the articles in each candidate sequence according to a sequence evaluation model, then calculating each candidate sequence according to the score value, and scoring again, wherein the sequence with the highest score is the optimal sequence.
Alternatively, the evaluation model may be modeled by means of RNN, as shown in fig. 4. The neural nodes in the graph are sequentially expanded from left to right in time sequence, x is an input layer, h is a hidden layer, and y is an output layer. h ist-n，……,，ht-2，ht-1，htNeural nodes, x, representing hidden statest-n,……,xt-2,xt-1,xtFeatures, y, representing articles in the sequencet-n,……,yt-2,yt-1,ytIndicating the probability of whether each article in the sequence will be clicked on.
In this example, the RNN network may be trained in advance. A large amount of page article data that a user requests to refresh can be utilized as a training sample. Each sample includes a sequence of articles presented to the user and a probability of clicking on each article in the sequence. The article clicked by the user in the article sequence display process has the click probability of 1; for the article which is not clicked by the user in the showing process of the article sequence, the clicking probability is 0. The training process of the RNN network may employ a classical RNN training method, such as BPTT or the like.
In the prediction stage, for each candidate sequence input, the evaluation model can predict the click probability of each article in the sequence. Then, the server adds all the click probabilities predicted in the candidate sequence to obtain the total score in the candidate sequence.
It is noted that the hidden layer of the RNN includes self-join weights W between hidden state neural networks, which can join features of adjacent articles in the sequence. Thus, when features of the recommendation information are put into the RNN network in order for learning, the RNN can learn the connection between adjacent articles through update iteration of W. When predicting the click probability of an article, the click probability of the article is not only independently predicted based on the characteristics of the article itself, but is predicted by contacting the characteristics of other articles in the sequence. The overall scoring of the candidate article sequence may then take into account the characteristics of each article itself and the association between each article. Thus, sequences selected from the candidate article sequences based on the overall ranking of the sequences tend to be superior to the initial article sequence.
(4) Issuing the refreshing article
In this step, the sequences corresponding to the n articles arranged in the front are intercepted from the selected optimal article sequence as the article sequence issued this time.
In practical applications, the operations described above are still executed when the next refresh is performed, and a new article sequence is obtained for issuing.
As shown in fig. 6, comparing the new method provided by the present application example with the original method, it can be known that: in the application example, in the article sequence issued by the current refreshing, the article sequence possibly issued by the future refreshing is considered, and the optimal sequence of the current refreshing is displayed for the user, and the current refreshing and the future refreshing are combined together to be displayed to be the integral optimal sequence. Therefore, in the continuous refreshing process of the user, the displayed content is refreshed to attract the user, and the user obtains better browsing experience.
Referring to fig. 7, an embodiment of the present invention further provides a device for refreshing recommendation information, including:
an initial sequence generating module 100, configured to respond to a refresh request from a request end, extract recommendation information from a recommendation information set, and combine the recommendation information into an initial information sequence; wherein the initial information sequence comprises M pieces of recommendation information;
the optimal sequence generation module 200 is configured to adjust an arrangement sequence of the initial information sequence according to the predicted click probability of each piece of recommended information in the initial information sequence, so as to obtain an optimal information sequence;
a recommended sequence extracting module 300, configured to extract a recommended information sequence from the preferred information sequence, where the recommended information sequence includes top N pieces of recommended information in the preferred information sequence, where M is greater than N; and
a sequence sending module 400, configured to send the recommendation information sequence to the request end, so as to perform refresh display at the request end.
In one possible implementation, referring to fig. 8, the initial sequence generating module 100 may include:
a recommendation information search unit 110, configured to search, according to the user information in the refresh request, corresponding recommendation information from a recommendation information base to obtain a recommendation information set;
an information weight calculating unit 120, configured to calculate a recommendation weight value of each piece of recommendation information in the recommendation information set according to a preset recommendation algorithm; and
a recommendation information sorting unit 130, configured to sort recommendation information in the recommendation information set according to a recommendation weight value; and
the recommendation information selecting unit 140 is configured to select the top M pieces of recommendation information, and combine them into an initial information sequence.
In one possible implementation manner, referring to fig. 9, the preferred sequence generating module 200 may include:
a candidate sequence generating unit 210, configured to randomly adjust an arrangement order of the recommendation information in the initial information sequence to obtain a candidate information sequence;
a click probability calculation unit 220, configured to calculate, through a sequence evaluation model, a predicted click probability of each piece of recommended information of the candidate information sequence;
a recommendation score calculating unit 230, configured to calculate recommendation scores of the candidate information sequences according to predicted click probabilities of pieces of recommendation information of the candidate information sequences; and
and a sequence selecting unit 240, configured to select, from the obtained candidate information sequences, a candidate information sequence with a highest recommendation score as a preferred information sequence.
In a possible implementation manner, the candidate sequence generating unit 210 is specifically configured to: performing variation operation on the arrangement positions of the recommended information in the initial information sequence to obtain a candidate information sequence; wherein the mutation operation comprises randomly selecting two pieces of recommended information from the initial information sequence for position exchange.
In a possible implementation manner, referring to fig. 10, the refresh apparatus may further include:
a training data module 500, configured to obtain training data of the sequence evaluation model; the training data comprises a sample recommendation information sequence of the request terminal and the click probability of each recommendation information in the sample recommendation information sequence; and
and a model training module 600, configured to input the training data into the sequence evaluation model according to the generation time of the sample recommendation information sequence, so as to train and update the sequence evaluation model.
The functions of the device can be realized by hardware, and can also be realized by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above-described functions.
The functions of the device can be realized by hardware, and can also be realized by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above-described functions.
In one possible design, the recommendation information refreshing structure includes a processor and a memory, the memory is used for the recommendation information refreshing apparatus to execute the recommendation information refreshing program in the first aspect, and the processor is configured to execute the program stored in the memory. The device for refreshing the recommendation information may further include a communication interface, and the device for refreshing the recommendation information may communicate with other devices or a communication network.
An embodiment of the present invention further provides a terminal device for refreshing recommendation information, as shown in fig. 11, where the terminal device includes: a memory 21 and a processor 22, the memory 21 having stored therein computer programs that may be executed on the processor 22. The processor 22 implements the method of refreshing recommendation information in the above-described embodiments when executing the computer program. The number of the memory 21 and the processor 22 may be one or more.
The apparatus further comprises:
a communication interface 23 for communication between the processor 22 and an external device.
The memory 21 may comprise a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.
If the memory 21, the processor 22 and the communication interface 23 are implemented independently, the memory 21, the processor 22 and the communication interface 23 may be connected to each other through a bus and perform communication with each other. The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown in FIG. 11, but this is not intended to represent only one bus or type of bus.
Optionally, in a specific implementation, if the memory 21, the processor 22 and the communication interface 23 are integrated on a chip, the memory 21, the processor 22 and the communication interface 23 may complete mutual communication through an internal interface.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and alternate implementations are included within the scope of the preferred embodiment of the present invention in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present invention.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
The computer readable media of embodiments of the present invention may be computer readable signal media or computer readable storage media or any combination of the two. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable read-only memory (CDROM). Additionally, the computer-readable storage medium may even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
In embodiments of the present invention, a computer readable signal medium may comprise a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, input method, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, Radio Frequency (RF), etc., or any suitable combination of the preceding.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
It will be understood by those skilled in the art that all or part of the steps carried by the method for implementing the above embodiments are programs that can be executed by associated hardware through instructions of the programs, and the programs can be stored in a computer readable storage medium, and when executed, comprise one or a combination of the steps of the method embodiments.
In addition, functional units in the embodiments of the present invention may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a separate product, may also be stored in a computer readable storage medium. The storage medium may be a read-only memory, a magnetic or optical disk, or the like.
While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A method for refreshing recommendation information is characterized by comprising the following steps:
responding to a refreshing request of a request terminal, extracting recommendation information from the recommendation information set to combine into an initial information sequence; wherein the initial information sequence comprises M pieces of recommendation information;
adjusting the arrangement sequence of the initial information sequence according to the predicted click probability of each piece of recommended information in the initial information sequence to obtain an optimal information sequence;
extracting a recommendation information sequence from the preference information sequence, wherein the recommendation information sequence comprises top N pieces of recommendation information in the preference information sequence, and M is greater than N; and
sending the recommended information sequence to the request end to refresh and display at the request end,
wherein, the adjusting the arrangement sequence of the initial information sequence according to the predicted click probability of each piece of recommended information in the initial information sequence to obtain an optimal information sequence includes:
randomly adjusting the arrangement sequence of the recommended information in the initial information sequence to obtain a candidate information sequence;
calculating the predicted click probability of each piece of recommended information of the candidate information sequence through a sequence evaluation model;
calculating recommendation scores of the candidate information sequences according to the predicted click probability of each piece of recommendation information of the candidate information sequences; and
and selecting the candidate information sequence with the highest recommendation score from the obtained candidate information sequences as a preferred information sequence.
2. The method for refreshing recommended information according to claim 1, wherein the step of extracting recommended information from the recommended information set in response to the request for refreshing from the requesting end and combining the extracted recommended information into an initial information sequence comprises:
searching corresponding recommendation information from a recommendation information base according to the user information in the refreshing request to obtain a recommendation information set;
calculating recommendation weight values of all recommendation information in the recommendation information set according to a preset recommendation algorithm; and
sorting all recommendation information in the recommendation information set according to recommendation weight values; and
and selecting the top M pieces of recommended information to combine into an initial information sequence.
3. The method for refreshing recommended information according to claim 1, wherein the randomly adjusting the ranking order of the recommended information in the initial information sequence to obtain the candidate information sequence comprises:
performing variation operation on the arrangement positions of the recommended information in the initial information sequence to obtain a candidate information sequence; wherein the mutation operation comprises randomly selecting two pieces of recommended information from the initial information sequence for position exchange.
4. The method for refreshing recommendation information according to claim 1, wherein said refreshing method further comprises:
acquiring training data of the sequence evaluation model; the training data comprises a sample recommendation information sequence of the request terminal and the click probability of each recommendation information in the sample recommendation information sequence; and
and inputting the training data into the sequence evaluation model according to the generation time of the sample recommendation information sequence so as to train and update the sequence evaluation model.
5. An apparatus for refreshing recommended information, comprising:
the initial sequence generation module is used for responding to a refreshing request of a request terminal and extracting recommendation information from the recommendation information set to combine the recommendation information into an initial information sequence; wherein the initial information sequence comprises M pieces of recommendation information;
the optimal sequence generation module is used for adjusting the arrangement sequence of the initial information sequence according to the predicted click probability of each piece of recommended information in the initial information sequence to obtain an optimal information sequence;
a recommended sequence extraction module, configured to extract a recommended information sequence from the preferred information sequence, where the recommended information sequence includes top N pieces of recommended information in the preferred information sequence, where M is greater than N; and
a sequence sending module, configured to send the recommendation information sequence to the request end to perform refresh display at the request end,
wherein the preferred sequence generation module comprises:
a candidate sequence generating unit, configured to randomly adjust an arrangement order of the recommendation information in the initial information sequence to obtain a candidate information sequence;
the click probability calculation unit is used for calculating the predicted click probability of each piece of recommended information of the candidate information sequence through a sequence evaluation model;
the recommendation score calculating unit is used for calculating recommendation scores of the candidate information sequences according to the predicted click probability of each piece of recommendation information of the candidate information sequences; and
and the sequence selection unit is used for selecting the candidate information sequence with the highest recommendation score from the obtained candidate information sequences as the preferred information sequence.
6. The apparatus for refreshing recommended information according to claim 5, wherein the initial sequence generating module comprises:
the recommendation information searching unit is used for searching corresponding recommendation information from a recommendation information base according to the user information in the refreshing request to obtain a recommendation information set;
the information weight calculation unit is used for calculating recommendation weight values of all recommendation information in the recommendation information set according to a preset recommendation algorithm; and
the recommendation information sorting unit is used for sorting the recommendation information in the recommendation information set according to a recommendation weight value; and
and the recommendation information selecting unit is used for selecting the recommendation information which is sequenced at the top M and combining the recommendation information into an initial information sequence.
7. The apparatus for refreshing recommendation information according to claim 5, wherein the candidate sequence generating unit is specifically configured to: performing variation operation on the arrangement positions of the recommended information in the initial information sequence to obtain a candidate information sequence; wherein the mutation operation comprises randomly selecting two pieces of recommended information from the initial information sequence for position exchange.
8. The apparatus for refreshing recommended information according to claim 5, wherein the refreshing apparatus further comprises:
the training data module is used for acquiring training data of the sequence evaluation model; the training data comprises a sample recommendation information sequence of the request terminal and the click probability of each recommendation information in the sample recommendation information sequence; and
and the model training module is used for inputting the training data into the sequence evaluation model according to the generation time of the sample recommendation information sequence so as to train and update the sequence evaluation model.
9. A refreshing terminal device for realizing recommendation information is characterized in that the terminal device comprises:
one or more processors;
storage means for storing one or more programs;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of refreshing recommendation information of any of claims 1-4.
10. A computer-readable storage medium storing a computer program, wherein the program, when executed by a processor, implements the method of refreshing recommendation information according to any one of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810846642.1A CN109063104B (en) | 2018-07-27 | 2018-07-27 | Recommendation information refreshing method and device, storage medium and terminal equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810846642.1A CN109063104B (en) | 2018-07-27 | 2018-07-27 | Recommendation information refreshing method and device, storage medium and terminal equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109063104A CN109063104A (en) | 2018-12-21 |
| CN109063104B true CN109063104B (en) | 2020-11-10 |
Family
ID=64836690
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810846642.1A Active CN109063104B (en) | 2018-07-27 | 2018-07-27 | Recommendation information refreshing method and device, storage medium and terminal equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109063104B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670117B (en) * | 2018-12-28 | 2023-03-31 | 北京百度网讯科技有限公司 | Information list recommendation method and device |
| CN111382346B (en) * | 2018-12-28 | 2023-09-01 | 第四范式（北京）技术有限公司 | Method and system for recommending content |
| CN111800446B (en) * | 2019-04-12 | 2023-11-07 | 北京沃东天骏信息技术有限公司 | Scheduling processing method, device, equipment and storage medium |
| CN112860985A (en) * | 2019-11-27 | 2021-05-28 | 云闪信(珠海市横琴)信息科技有限公司 | Application navigation method and device, electronic equipment and computer readable storage medium |
| CN112732927A (en) * | 2020-12-31 | 2021-04-30 | 平安资产管理有限责任公司 | Content similarity analysis method and device based on knowledge graph |
| CN113190758B (en) * | 2021-05-21 | 2023-01-20 | 聚好看科技股份有限公司 | Server and media asset recommendation method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123284A (en) * | 2013-04-24 | 2014-10-29 | 华为技术有限公司 | Recommendation method and server |
| CN104899195A (en) * | 2014-01-26 | 2015-09-09 | 武汉联影医疗科技有限公司 | Customized educational resource recommending method and apparatus |
| CN107463704A (en) * | 2017-08-16 | 2017-12-12 | 北京百度网讯科技有限公司 | Searching method and device based on artificial intelligence |
| CN107977411A (en) * | 2017-11-21 | 2018-05-01 | 腾讯科技(成都)有限公司 | Group recommending method, device, storage medium and server |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20140102381A (en) * | 2013-02-13 | 2014-08-22 | 삼성전자주식회사 | Electronic device and Method for recommandation contents thereof |
-
2018
- 2018-07-27 CN CN201810846642.1A patent/CN109063104B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123284A (en) * | 2013-04-24 | 2014-10-29 | 华为技术有限公司 | Recommendation method and server |
| CN104899195A (en) * | 2014-01-26 | 2015-09-09 | 武汉联影医疗科技有限公司 | Customized educational resource recommending method and apparatus |
| CN107463704A (en) * | 2017-08-16 | 2017-12-12 | 北京百度网讯科技有限公司 | Searching method and device based on artificial intelligence |
| CN107977411A (en) * | 2017-11-21 | 2018-05-01 | 腾讯科技(成都)有限公司 | Group recommending method, device, storage medium and server |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109063104A (en) | 2018-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109063104B (en) | Recommendation information refreshing method and device, storage medium and terminal equipment | |
| CN109902708B (en) | Recommendation model training method and related device | |
| CN110020094B (en) | Display method and related device for search results | |
| RU2725659C2 (en) | Method and system for evaluating data on user-element interactions | |
| US11294891B2 (en) | Interactive search experience using machine learning | |
| CN110309405B (en) | Project recommendation method and device and storage medium | |
| EP4181026A1 (en) | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium | |
| CN112307344B (en) | Object recommendation model, object recommendation method and device and electronic equipment | |
| CN111506820B (en) | Recommendation model, recommendation method, recommendation device, recommendation equipment and recommendation storage medium | |
| Brodén et al. | Ensemble recommendations via thompson sampling: an experimental study within e-commerce | |
| CN111506821A (en) | Recommendation model, method, device, equipment and storage medium | |
| CN112883265A (en) | Information recommendation method and device, server and computer readable storage medium | |
| CN110598084A (en) | Object sorting method, commodity sorting device and electronic equipment | |
| WO2023185925A1 (en) | Data processing method and related apparatus | |
| CN110851651A (en) | Personalized video recommendation method and system | |
| CN114896517A (en) | Commodity recommendation method, system, equipment and storage medium | |
| CN111489196B (en) | Prediction method and device based on deep learning network, electronic equipment and medium | |
| CN112000326A (en) | Service providing method and device, electronic equipment and storage medium | |
| CN111159558B (en) | Recommendation list generation method and device and electronic equipment | |
| CN113901056A (en) | Interface recommendation method and device and electronic equipment | |
| CN113191840A (en) | Article information display method and device, electronic equipment and computer readable medium | |
| CN113450167A (en) | Commodity recommendation method and device | |
| CN117391824B (en) | Method and device for recommending articles based on large language model and search engine | |
| CN112905885B (en) | Method, apparatus, device, medium and program product for recommending resources to user | |
| CN112949852B (en) | Method and device for training model, electronic equipment and readable storage medium |
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 |