CN106777067A - Information recommendation method and system - Google Patents

Abstract

The present invention provides information recommendation method and system, including：The multiple sampled data streams read according to prefixed time interval cutting are cut into slices with obtaining multiple sampled datas；User behavior scene is defined, the user behavior scene includes at least one and/or at least one class user behavior event；The part for meeting user behavior scene is selected out to constitute scene snapshot from the data of each sampled data section respectively；Using scene snapshot as scenario simulation model input, so as to obtain scenario simulation model output scene portrait；Scene portrait is matched with default recommendation results, if the match is successful, default recommendation results is defined as recommendation information.The present invention realizes the dynamically interactive type personalized recommendation of stream data reading, realizes reading in sampled data, instant personalized recommendation immediately based on user behavior data, while ensureing ageing, the individual individual demand of different user can be taken into account again.

Description

Information recommendation method and system

technical field

The invention relates to the field of information recommendation, in particular to an information recommendation method and system.

Background technique

Streaming information services with a wide social coverage and a large number of active online users are called public streaming information services, and the typical representatives of public streaming information services are various media live broadcast services. The recommendation of public streaming information services faces the following problems: 1) It must be recommended within the online cycle of the content, so as to avoid recommending offline content to users; 2) It must be recommended to appropriate individual users, so as to avoid other users Groups interfere.

It can be seen that for the recommendation of public streaming information services, a new technology that can ensure timeliness and take into account the personalized needs of different individual users is needed.

Contents of the invention

In view of the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide an information recommendation method and system, which is used to solve the problem that the recommendation of public streaming information services in the prior art cannot guarantee timeliness and cannot take into account the personalities of different individual users needs and other issues.

In order to achieve the above purpose and other related purposes, the present invention provides an information recommendation method, including: cutting multiple sampled data streams read in according to preset time intervals to obtain multiple sampled data slices; defining one or more user behavior scenarios , each user behavior scenario includes at least one and/or at least one type of user behavior event; respectively select a part that conforms to the user behavior scenario from the data of each sampled data slice to form a scene snapshot; The scene snapshot is used as the input of the scene simulation model, thereby obtaining the scene portrait output by the scene simulation model; the scene portrait is matched with the preset recommendation result, and if the matching is successful, the preset recommendation result is determined as a recommendation information.

In an embodiment of the present invention, when one or more user behavior scenarios are defined, the one or more user behavior scenarios constitute a behavior pattern, and the scene portrait of a certain user behavior scenario in the behavior pattern is determined by the The scene snapshot of a certain user behavior scene and the scene portrait of the previous user behavior scene are jointly used as the input of the scene simulation model, and are obtained through the output of the scene simulation model, and the method also includes: the final obtained The scene portrait is matched with the preset recommendation result, and if the matching is successful, the preset recommendation result is determined as recommendation information.

In an embodiment of the present invention, the scene simulation model includes a plurality, each of which is used to perform scene simulation for a kind of feature information.

In an embodiment of the present invention, the scene simulation model includes: a supervised learning model, and/or an unsupervised learning model.

In an embodiment of the present invention, the unsupervised learning model includes: a deep learning model.

In an embodiment of the present invention, the deep learning model is established according to the backpropagation algorithm.

In an embodiment of the present invention, each of the sampled data streams includes a type of user behavior event data, and the plurality of sampled data slices include a plurality of user behavior events corresponding to the plurality of sampled data streams respectively. Partial data of the event.

In an embodiment of the present invention, each type of user behavior event data includes a plurality of user behavior event data, and each user behavior event data includes: a time stamp, an ID for identifying the object that generated the corresponding behavior event Sampling data of user identification and corresponding behavior events.

In an embodiment of the present invention, before inputting the scene snapshot into the scene simulation model, the method further includes: sorting the user behavior event data included in the scene snapshot in chronological order, and inputting the scene snapshot in sequence The scenario simulation model described above.

In an embodiment of the present invention, the preset recommendation results include multiple ones, and the matching includes: respectively calculating the correlation between the scene portrait and each of the preset recommendation results, and selecting the preset recommendation result with the greatest correlation Let the recommendation result be determined as the recommendation information.

In order to achieve the above purpose and other related purposes, the present invention provides an information recommendation system, including: a data slicing module, which is used to cut and read multiple sampled data streams according to preset time intervals to obtain multiple sampled data slices; scene definition A module for defining one or more user behavior scenarios, each user behavior scenario including at least one and/or at least one type of user behavior event; a snapshot selection module for selecting from the data of each sampled data slice respectively Parts that meet the user behavior scene to form a scene snapshot; the scene simulation module is used to use the scene snapshot as the input of the scene simulation model, so as to obtain the scene portrait output by the scene simulation model; the matching module is used to The scene portrait is matched with a preset recommendation result, and if the match is successful, the preset recommendation result is determined as recommendation information.

In an embodiment of the present invention, when one or more user behavior scenarios are defined, the one or more user behavior scenarios constitute a behavior pattern, and the scene portrait of a certain user behavior scenario in the behavior pattern is determined by the The scene snapshot of a certain user behavior scene and the scene portrait of the previous user behavior scene are used as the input of the scene simulation model, and are obtained through the output of the scene simulation model. The matching module is also used for: the final The obtained scene portrait is matched with the preset recommendation result, and if the matching is successful, the preset recommendation result is determined as recommendation information.

In an embodiment of the present invention, the scene simulation model includes a plurality, each of which is used to perform scene simulation for a kind of feature information.

In an embodiment of the present invention, the scene simulation model includes: a supervised learning model, and/or an unsupervised learning model.

In an embodiment of the present invention, the unsupervised learning model includes: a deep learning model.

In an embodiment of the present invention, the deep learning model is established according to the backpropagation algorithm.

In an embodiment of the present invention, each of the sampled data streams includes a type of user behavior event data, and the plurality of sampled data slices include a plurality of user behavior events corresponding to the plurality of sampled data streams respectively. Partial data of the event.

In an embodiment of the present invention, each type of user behavior event data includes a plurality of user behavior event data, and each user behavior event data includes: a time stamp, an ID for identifying the object that generated the corresponding behavior event Sampling data of user identification and corresponding behavior events.

In an embodiment of the present invention, the system further includes: a sorting module, configured to sort the user behavior event data contained in the scene snapshot in chronological order before inputting the scene snapshot into the scene simulation model, And sequentially input the scene simulation model.

In an embodiment of the present invention, the preset recommendation results include multiple ones, and the matching includes: respectively calculating the correlation between the scene portrait and each of the preset recommendation results, and selecting the preset recommendation result with the greatest correlation Let the recommendation result be determined as the recommendation information.

As mentioned above, the information recommendation method and system of the present invention realizes the technical solution of real-time reading of sampled data and real-time personalized recommendation based on user behavior data, and has the following beneficial effects:

1) The time period for updating the recommendation results is short, calculated in minutes or seconds or smaller time units;

2) After the sampling data is read in, the impact of the sampling data on the recommendation results will appear in a few minutes/seconds, and an interactive recommendation of "behavior sampling-recommendation feedback" can be formed;

3) The recommendation result is based on user behavior data sampling, which is the probability result of the historical evolution of individual user behavior and has personalized characteristics.

Description of drawings

FIG. 1 is a flowchart of an information recommendation method according to an embodiment of the present invention.

FIG. 2 is a schematic diagram of cutting a sampled data stream to generate sampled data slices according to an embodiment of the present invention.

FIG. 3 is a schematic diagram showing the internal composition of a type A behavior sampling data stream according to an embodiment of the present invention.

Fig. 4 is a schematic diagram of a defined user behavior scenario according to an embodiment of the present invention.

5A-5B are schematic diagrams showing a snapshot selection process according to an embodiment of the present invention.

6A-6B are schematic diagrams of a scene simulation process according to an embodiment of the present invention.

FIG. 7 is a schematic diagram of a simulation process in which multiple user behavior scenarios are connected in series to form a behavior pattern according to an embodiment of the present invention.

FIG. 8 is a schematic diagram of a process of selecting a final recommendation result according to an embodiment of the present invention.

FIG. 9 is a block diagram of an information recommendation system according to an embodiment of the present invention.

detailed description

Embodiments of the present invention are described below through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation modes, and various modifications or changes can be made to the details in this specification based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, in the case of no conflict, the following embodiments and features in the embodiments can be combined with each other.

It should be noted that the diagrams provided in the following embodiments are only schematically illustrating the basic ideas of the present invention, and only the components related to the present invention are shown in the diagrams rather than the number, shape and shape of the components in actual implementation. Dimensional drawing, the type, quantity and proportion of each component can be changed arbitrarily during actual implementation, and the component layout type may also be more complicated.

With the advent of the era of big data, the accuracy of information recommendation can only be improved by mining the historical behavior of users to predict the future behavior of users and establishing the relationship between users and content. Please refer to Fig. 1, the present invention provides an information recommendation method, on the basis of ensuring timeliness, taking into account the personalized needs of different users, and realizing the dynamic interactive personalized recommendation of streaming data reading, specifically including the following steps:

Step S101: Cut the read-in multiple sampling data streams according to preset time intervals to obtain multiple sampling data slices. Figure 2 shows a cutting process, and the cutting time point is recorded as time point 1, and each sampling data The stream includes a type of user behavior event data, and each sliced sampled data slice includes a part corresponding to a preset time interval in the corresponding type of user behavior event data. As shown in Figure 3, each sampled data stream is composed of multiple user behavior event data, where each user behavior event data includes: a timestamp recording the time when the data was generated, and the user identifying the source object of the data generation identification, and the data subject section.

It should be noted that in each sampled data stream, each user behavior event data may be out of order. When cutting at a preset time interval, the user behavior event data corresponding to the time stamp within the preset time interval range are segmented to generate sample data slices of this type of user behavior events. After multiple cuts, the batches of sampling data slices are sorted in the order of cutting time.

For example, the user behavior event is: viewers watch a certain channel, and the corresponding sampling data stream is: the data of all online viewers watching this channel, these data will be obtained in the form of data stream in real time, and the corresponding sampling data slice is: in the time axis After performing multiple slice processing on these data, a set of slices sorted by cutting time are obtained.

Step S102: Define one or more user behavior scenarios. As shown in FIG. 4, each defined user behavior scenario may include one, multiple, one type, or multiple types of user behavior events.

For example: Type A user behavior event is: the audience watches a certain channel, type B user behavior event is: the audience switches to a certain channel, and type C user behavior event is: the channel plays a certain program, when the defined scene a includes two types of A and C , then scene a is: the audience watches a certain program on a certain channel, and when the defined scene b includes two types of B and C, then scene b is: the audience switches to a certain program on a certain channel.

Step S103: For each cut, select a part that conforms to the user behavior scene from the data of each sampled data slice to form a scene snapshot corresponding to each cut time point.

For example, for Figure 5A, the sampled data slice of type A user behavior events is: the data of all online viewers watching a certain channel in different time periods, and after performing multiple slice processing on these data on the time axis, a group of Slices sorted by time. Type C user behavior event sampling data slices are: program list data of all live channels broadcasting programs within a certain period of time, after slicing these data three times on the time axis, the obtained groups are sorted according to the cutting time slices. The selected scene a snapshot 1 is: at time point 1, the audience watches a snapshot composed of sampled data of a program on a certain channel, and the selected scene a snapshot 2 is: at time point 2, the audience watches a snapshot composed of sampled data of a certain program on a certain channel Snapshot, the selected scene a snapshot 3 is: at time point 3, a snapshot composed of sampling data of a program watched by a certain channel. For Figure 5B, the sampled data slice of B-type user behavior events is: the data of all online viewers switching to a certain channel in different time periods, and after processing these data slices three times on the time axis, a set of data obtained according to each other is obtained. Time-sorted slices. Type C user behavior event sampling data slices are: the program list data of all live channels broadcasting programs within a certain period of time. After performing multiple slice processing on these data on the time axis, a set of time-ordered data is obtained. slice. The selected scene b snapshot 1 is: time point 1, a snapshot composed of sampled data switched in by viewers of a certain channel and a certain program, and the selected scene b snapshot 2 is: time point 2, the sampled data of a certain channel and certain program viewers switched in The selected snapshot 3 of scene b is: at time point 3, a snapshot composed of sampled data switched in by viewers of a certain program on a certain channel.

Step S104: According to the order of cutting time, each of the scene snapshots is used as the input of the scene simulation model, so as to obtain the scene portraits output by the scene simulation model. It should be noted that the scene simulation model is a customized model, and there may be one or more than one model, and each model is used to perform scene simulation for a kind of feature information. The generation of the custom model is based on the needs of the portrait, and is composed of various selected deep learning algorithms (such as: backpropagation algorithm) and/or unsupervised learning algorithms.

For example, for FIG. 6A , the simulation model simulates the probability profile of a specific feature contained in the snapshot of scene a. Specifically, the scene a portrait 1 is: the probability portrait of the specific feature contained in the scene a snapshot 1 simulated, that is, the probability prediction that the audience may watch a certain program on a certain channel at time point 1 and its adjacent time domain, The scene a portrait 1 is compared with the scene a snapshot 1 and the parameters are adjusted to reduce the difference between the two; the scene a portrait 2 is: the probability portrait of the feature simulated for the specific feature contained in the scene a snapshot 2, that is, the time Point 2 and its adjacent time domain, the probability prediction that the audience may watch a certain program on a certain channel, compare the scene a portrait 2 with the scene a snapshot 2 and adjust the parameters, so that the difference between the two is reduced; the scene a portrait 3 is: for The probability profile of the feature simulated by the specific features contained in scene a snapshot 3, that is, time point 3 and its adjacent time domain, the probability prediction that the audience may watch a certain program on a certain channel, scene a portrait 3 and scene a snapshot 3 Compare and adjust the parameters to reduce the difference between the two. By analogy, the scene a portrait n is: the probability portrait of the feature simulated for the specific feature contained in the scene a snapshot n, that is, the probability prediction that the audience may watch a certain program on a certain channel at time point n and its adjacent time domain , the scene a portrait n is compared with the scene a snapshot n and the parameters are adjusted so that the difference between the two is reduced.

For another example, for FIG. 6B , the simulation model simulates the probability profile of a specific feature contained in the snapshot of scene b. Specifically, the scene b portrait 1 is: the probability portrait of the specific feature contained in the scene b snapshot 1 simulated, that is, the probability prediction that the audience may switch to a certain channel and a certain program at time point 1 and its adjacent time domain , the scene b portrait 1 is compared with the scene b snapshot 1 and the parameters are adjusted to reduce the difference between the two; the scene b portrait 2 is: the probability portrait of the feature simulated for the specific feature contained in the scene b snapshot 2, that is At time point 2 and its adjacent time domain, the probability prediction that the audience may switch to a certain program on a certain channel, the scene b portrait 1 is compared with the scene b snapshot 1 and the parameters are adjusted to reduce the difference between the two; the scene b portrait 3 is : The probability portrait of the feature simulated for the specific feature contained in the snapshot 3 of scene b, that is, the probability prediction that the audience may switch to a certain channel and a certain program at time point 3 and its adjacent time domain, scene b portrait 1 and scene b Snapshot 1 is compared and parameters are adjusted to reduce the difference between the two. By analogy, the scene b profile n is: the probability profile of the feature simulated for the specific feature contained in the scene b snapshot n, that is, the probability prediction that the audience may watch a certain program on a certain channel at time point n and its adjacent time domain , the scene b portrait n is compared with the scene b snapshot n and the parameters are adjusted so that the difference between the two is reduced.

It should be noted that, in another embodiment, before each scene snapshot is input into the corresponding scene simulation model, according to the timestamp of the user behavior event, the user behavior event data contained in the scene snapshot are firstly sorted in chronological order sorted, and then input into the corresponding scene simulation model in sequence.

In particular, when one or more user behavior scenarios are defined, the one or more user behavior scenarios constitute a behavior pattern, specifically, a single user behavior scenario may constitute a behavior pattern, and multiple user behavior scenarios are combined ( For example in series) can also constitute behavioral patterns. Figure 7 shows a situation in which multiple user behavior scenarios are connected in series to form a behavior pattern, wherein the scene portrait of a certain user behavior scenario in the behavior pattern is composed of the scene snapshot of the certain user behavior scenario and the previous user behavior scenario The scene portraits are used as the input of the scene simulation model and obtained through the output of the scene simulation model.

For example, two factors are considered at the same time: 1) at time point 1 and its adjacent time domain, the probability that the audience may watch a certain program on a certain channel comes from scene a portrait 1; 2) at time point 1, the audience switches to a certain channel and certain program Actual sample data for the show, from scene b snapshot 1. Simulation: time point 1 and its adjacent time domain, the probability prediction that viewers may switch to a certain program on a certain channel.

Step S105: Match the scene portrait with a preset recommendation result, and if the matching is successful, determine the preset recommendation result as recommendation information. Optionally, the preset recommendation results may include multiple ones, and the matching specifically includes: separately calculating the correlation between each scene portrait and each of the preset recommendation results, and determining the preset recommendation result with the greatest correlation For the final recommendation information.

For example, as shown in Figure 8, the scene b portrait output by the behavior pattern simulation composed of all the scene simulations is used as the final recommendation result, then, the audience can switch to a certain channel according to a certain time point and its adjacent time domain The probability of a program is used to recommend channels to viewers.

Please refer to FIG. 9 , similar to the principles of the above-mentioned method embodiments, the present invention provides an information recommendation system 900, which is implemented as a software and can be carried on an electronic device with input, output, and data processing functions for execution. Since each technical feature in the preceding embodiments can be applied to this system embodiment, it is not repeated here.

The system 900 includes: a data slicing module 901 , a scene definition module 902 , a snapshot selection module 903 , a scene simulation module 904 , and a matching module 905 . Specifically: the data slicing module 901 cuts the read-in multiple sampling data streams according to preset time intervals to obtain multiple sampling data slices; the scene definition module 902 defines one or more user behavior scenarios, each user behavior scenario includes at least one And/or at least one type of user behavior event; the snapshot selection module 903 selects the part that meets the user behavior scene from the data of each sampled data slice to form a scene snapshot; the scene simulation module 904 takes the scene snapshot As the input of the scene simulation model, the scene portrait output by the scene simulation model is obtained; the matching module 905 matches the scene portrait with the preset recommendation result, and if the matching is successful, the preset recommendation result is determined as the recommendation information.

In an embodiment, the system further includes: a sorting module, configured to sort the user behavior event data included in the scene snapshot in chronological order before inputting the scene snapshot into the scene simulation model, and sort input the scenario simulation model in sequence.

To sum up, the present invention effectively overcomes various shortcomings in the prior art and has high industrial application value.

The above-mentioned embodiments only illustrate the principles and effects of the present invention, but are not intended to limit the present invention. Anyone skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical ideas disclosed in the present invention should still be covered by the claims of the present invention.

Claims (20)

1. An information recommendation method, characterized in that, comprising: Cut the read-in multiple sampling data streams according to preset time intervals to obtain multiple sampling data slices; Define one or more user behavior scenarios, each user behavior scenario includes at least one and/or at least one type of user behavior event; Respectively select a part conforming to the user behavior scene from the data of each sampled data slice to form a scene snapshot; Using the scene snapshot as the input of the scene simulation model, thereby obtaining the scene portrait output by the scene simulation model; The scene portrait is matched with a preset recommendation result, and if the matching is successful, the preset recommendation result is determined as recommendation information. 2. The information recommendation method according to claim 1, wherein when one or more user behavior scenarios are defined, the one or more user behavior scenarios constitute a behavior pattern, and one of the behavior patterns The scene portrait of the user behavior scene is obtained by the scene snapshot of the certain user behavior scene and the scene portrait of the previous user behavior scene together as the input of the scene simulation model, and obtained through the output of the scene simulation model, so The method further includes: matching the finally obtained scene portrait with the preset recommendation result, and if the matching is successful, determining the preset recommendation result as recommendation information. 3 . The information recommendation method according to claim 1 , wherein the scenario simulation model includes multiple scenarios, each of which is used to perform scenario simulation for a kind of feature information. 4 . 4. The information recommendation method according to claim 1, wherein the scene simulation model comprises: a supervised learning model and/or an unsupervised learning model. 5. The information recommendation method according to claim 4, wherein the unsupervised learning model comprises: a deep learning model. 6. The information recommendation method according to claim 5, wherein the deep learning model is established according to a backpropagation algorithm. 7. The information recommendation method according to claim 1, wherein each of the sampled data streams includes a type of user behavior event data, and the plurality of sampled data slices include a data segment corresponding to the plurality of sampled data streams respectively. Partial data corresponding to multiple user behavior events. 8. The information recommendation method according to claim 7, wherein each type of user behavior event data includes a plurality of user behavior event data, and each of the user behavior event data includes: a timestamp, for Identify the user ID of the object that generates the corresponding behavior event, and the sampling data of the corresponding behavior event. 9. The information recommendation method according to claim 8, further comprising: prior to inputting the scene snapshot into the scene simulation model, sorting the user behavior event data contained in the scene snapshot in chronological order , and input the scene simulation model in sequence. 10. The information recommendation method according to claim 1, wherein the preset recommendation results include multiple, and the matching includes: respectively calculating the correlation between the scene portrait and each of the preset recommendation results , determining the preset recommendation result with the highest correlation as the recommendation information. 11. An information recommendation system, characterized in that it comprises: A data slicing module, configured to cut and read multiple sampled data streams according to preset time intervals to obtain multiple sampled data slices; A scenario definition module, configured to define one or more user behavior scenarios, each user behavior scenario includes at least one and/or at least one type of user behavior event; A snapshot selection module, configured to select a part conforming to the user behavior scene from the data of each sampled data slice to form a scene snapshot; The scene simulation module is used to use the scene snapshot as the input of the scene simulation model, so as to obtain the scene portrait output by the scene simulation model; A matching module, configured to match the scene portrait with a preset recommendation result, and if the matching is successful, determine the preset recommendation result as recommendation information. 12. The information recommendation system according to claim 11, wherein when one or more user behavior scenarios are defined, the one or more user behavior scenarios constitute a behavior pattern, and one of the behavior patterns The scene portrait of the user behavior scene is obtained by the scene snapshot of the certain user behavior scene and the scene portrait of the previous user behavior scene together as the input of the scene simulation model, and obtained through the output of the scene simulation model, so The matching module is further configured to: match the finally obtained scene portrait with the preset recommendation result, and if the matching is successful, determine the preset recommendation result as recommendation information. 13. The information recommendation system according to claim 11, wherein the scene simulation model comprises a plurality, each of which is used to perform scene simulation for a kind of feature information. 14. The information recommendation system according to claim 11, wherein the scene simulation model comprises: a supervised learning model and/or an unsupervised learning model. 15. The information recommendation system according to claim 14, wherein the unsupervised learning model comprises: a deep learning model. 16. The information recommendation system according to claim 15, wherein the deep learning model is established according to a backpropagation algorithm. 17. The information recommendation system according to claim 11, wherein each of the sampled data streams includes a type of user behavior event data, and the multiple sampled data slices include a Partial data corresponding to multiple user behavior events. 18. The information recommendation system according to claim 17, wherein each type of user behavior event data includes a plurality of user behavior event data, and each of the user behavior event data includes: timestamp, for Identify the user ID of the object that generates the corresponding behavior event, and the sampling data of the corresponding behavior event. 19. The information recommendation system according to claim 18, further comprising: a sorting module, configured to sort each user behavior event included in the scene snapshot before inputting the scene snapshot into the scene simulation model The data are sorted in chronological order, and are sequentially input into the scenario simulation model. 20. The information recommendation system according to claim 11, wherein the preset recommendation results include multiple, and the matching includes: separately calculating the correlation between the scene portrait and each of the preset recommendation results , determining the preset recommendation result with the greatest correlation as the recommendation information.