Online ctr calibration method based on combination of double-layer model and multi-dimensional information
A multi-dimensional, model-based technology, applied in the field of recommendation systems, can solve problems such as the inability to achieve refined sample distinctions, and achieve the effect of avoiding model instability and realizing traffic benefits
Pending Publication Date: 2022-04-22
TIANYI ELECTRONICS COMMERCE
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
At present, the CTR calibration of the recommendation system is mostly carried out through statistical formulas, and the same transformation process is performed on all samples. It is impossible to realize the fine distinction of samples and adjust the calibration results according to the performance of samples in different time periods.
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment 1
[0046] Such as Figure 1-2 , the present invention provides an online CTR calibration method based on a two-layer model combined with multi-dimensional information, including the following:
[0047] 1. Estimation module:
[0048] (1) pctr model training
[0049] Extract user i advertisement j stored in hive in recent t0-t1 days (sample s ij ) corresponding users, advertisements, behaviors, contexts and other features (hereinafter briefly described with model features) use the depth model to carry out pctr model training, the present invention does not limit the depth model structure; the obtained pctr result can be regarded as a result closer to the real value, and the rest On this basis, the module provides more information to make the model obtain more accurate results;
[0050] (2) Training set offline pctr prediction:
[0051] For the above-mentioned user i advertisement j, the pctr model prediction is performed, and the result train pctr is obtained ij And passed to th...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses an online ctr calibration method based on combination of a double-layer model and multi-dimensional information, and the method comprises the following steps: obtaining a reference pctr result through an estimation module, carrying out the pctr division of the reference pctr result through a matching module on the basis, regarding the samples in the same pctr interval as the same distribution, and avoiding the model instability caused by the sample sparsity. The corresponding multi-dimensional information is generated by using different intervals, and the real-time information is obtained by the real-time module, so that the samples can be distinguished by effectively using the multi-dimensional information. And finally, a more accurate ctr calibration result under the double-layer model is obtained through the calibration module, so that the flow revenue maximization is realized.
Description
technical field [0001] The invention relates to the field of recommendation systems, in particular to an online CTR calibration method based on a two-layer model combined with multi-dimensional information. Background technique [0002] Ctr calibration is an essential link in the recommendation system. A good calibration method can increase the revenue brought by the traffic on the basis of ensuring clicks; a poor calibration method will misestimate the actual click rate of the traffic and affect the evaluation of the expected revenue. [0003] In the actual recommendation scenario, each item will have a corresponding revenue bid, and the recommendation platform will calculate the expected revenue based on ctr*bid to obtain the optimal display order. How to accurately evaluate the CTR of each material is an important link in the recommendation system. At present, the CTR calibration of the recommendation system is mostly carried out through the formula obtained from statist...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06Q30/02
CPCG06Q30/0244G06Q30/0246G06Q30/0272
Inventor 韩弘炀傅剑文陈心童章建森周文彬
Owner TIANYI ELECTRONICS COMMERCE