Traffic Shaping in E-Commercial Search Engine: Multi-Objective Online Welfare Maximization

Authors: Liucheng Sun, Chenwei Weng, Chengfu Huo, Weijun Ren, Guochuan Zhang, Xin Li574-581

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We also run experiments on both synthetic and real-world datasets to validate our algorithms. extensive experiments are conducted. The simulation results on a synthetic dataset validate the effectiveness of our proposed algorithm and the A/B testing results in the real-world production environment exhibit the effectiveness and practicability of the method.
Researcher Affiliation Collaboration Liucheng Sun1,2, Chenwei Weng1, Chengfu Huo1, Weijun Ren1, Guochuan Zhang2, Xin Li1 1Alibaba Group , 2Zhejiang University liucheng.slc@alibaba-inc.com, wengchenwei.pt@alibaba-inc.com, chengfu.huocf@alibaba-inc.com afei@alibaba-inc.com, zgc@zju.edu.cn, xin.l@alibaba-inc.com
Pseudocode Yes Algorithm 1: Greedy Traffic Splitting Algorithm, Algorithm 2: Probabilistic Balance Algorithm, Algorithm 3: Improved Traffic Splitting Algorithm
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository.
Open Datasets No The paper uses a "synthetic dataset" for simulation, which is generated, and a "real-world production environment" (Alibaba's e-commerce search engine) for A/B testing, which is proprietary. There is no indication of public availability or access for these datasets.
Dataset Splits No The paper describes a synthetic dataset generation process and A/B test group splitting, but it does not specify explicit training, validation, or test dataset splits (e.g., "80/10/10 split" or specific sample counts for each split).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications) used for running its experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers, such as programming languages, libraries, or frameworks.
Experiment Setup Yes For each query j, 200 randomly sampled items are retrieved for ranking, and bj is sampled from a uniform distribution U(3, 50). For each item-query pair (i, j), the relevance score rij is drawn from the beta distributions Beta(3, 2) and Beta(2, 3); the click probabilities are drawn from the uniform distributions U(0.1, 0.3) and U(0, 0.2); the conversion probabilities follow the uniform distributions U(0, 0.01) and U(0, 0.005) for mature and new items respectively... We set |Ic| = 1, 000 and |Is| = 1, 000... The target numbers of clicks are 18 and 2 for mature and new items respectively. We fix p1 = 0.9... we change the value of p2 from 0 to 0.1 with the step size 0.01 and p3 = 1 p1 p2. We set p1 = 0.96, p2 = 0.02, p3 = 0.02 and the experiment lasts for a week.