arXiv subjects

Ziqiang Cheng

Publications and source records attributed to Ziqiang Cheng.

7 recordsLinked to original sources

arXiv2026

Detoxification for LLM: From Dataset Itself

Existing detoxification methods for large language models mainly focus on post-training stage or inference time, while few tackle the source of toxicity, namely, the dataset itself. Such training-based or controllable decoding approaches cannot completely suppress the model's inherent toxicity, whereas detoxifying the pretraining dataset can fundamentally reduce the toxicity that the model learns during training. Hence, we attempt to detoxify directly on raw corpora with SoCD (Soft Contrastive Decoding), which guides an LLM to localize and rewrite toxic spans in raw data while preserving semantics, in our proposed HSPD (Hierarchical Semantic-Preserving Detoxification) pipeline, yielding a detoxified corpus that can drop-in replace the original for fine-tuning or other training. On GPT2-XL, HSPD attains state-of-the-art detoxification, reducing Toxicity Probability (TP) from 0.42 to 0.18 and Expected Maximum Toxicity (EMT) from 0.43 to 0.20. We further validate consistent best-in-class results on LLaMA2-7B, OPT-6.7B, and Falcon-7B. These findings show that semantics-preserving, corpus-level rewriting with HSPD effectively suppresses downstream toxicity while retaining data utility and allowing seamless source-level mitigation, thereby reducing the cost of later model behavior adjustment. (Code is available at: https://github.com/ntsw2001/data_detox_for_llm)

cs.CL↗

arXiv2024

A new type of simplified inverse Lax-Wendroff boundary treatment I: hyperbolic conservation laws

In this paper, we design a new kind of high order inverse Lax-Wendroff (ILW) boundary treatment for solving hyperbolic conservation laws with finite difference method on a Cartesian mesh. This new ILW method decomposes the construction of ghost point values near inflow boundary into two steps: interpolation and extrapolation. At first, we impose values of some artificial auxiliary points through a polynomial interpolating the interior points near the boundary. Then, we will construct a Hermite extrapolation based on those auxiliary point values and the spatial derivatives at boundary obtained via the ILW procedure. This polynomial will give us the approximation to the ghost point value. By an appropriate selection of those artificial auxiliary points, high-order accuracy and stable results can be achieved. Moreover, theoretical analysis indicates that comparing with the original ILW method, especially for higher order accuracy, the new proposed one would require fewer terms using the relatively complicated ILW procedure and thus improve computational efficiency on the premise of maintaining accuracy and stability. We perform numerical experiments on several benchmarks, including one- and two-dimensional scalar equations and systems. The robustness and efficiency of the proposed scheme is numerically verified.

math.NA↗

arXiv2021

How Powerful are Interest Diffusion on Purchasing Prediction: A Case Study of Taocode

A taocode is a kind of specially coded text-link on Taobao(the world's biggest online shopping website), through which users can share messages about products with each other. Analyzing taocodes can potentially facilitate understanding of the social relationships between users and, more excitingly, their online purchasing behaviors under the influence of taocode diffusion. This paper innovatively investigates the problem of online purchasing predictions from an information diffusion perspective, with taocode as a case study. Specifically, we conduct profound observational studies on a large-scale real-world dataset from Taobao, containing over 100M Taocode sharing records. Inspired by our observations, we propose InfNet, a dynamic GNN-based framework that models the information diffusion across Taocode. We then apply InfNet to item purchasing predictions. Extensive experiments on real-world datasets validate the effectiveness of InfNet compared with 8 state-of-the-art baselines.

cs.SI↗

arXiv2020

Time2Graph: Revisiting Time Series Modeling with Dynamic Shapelets

Time series modeling has attracted extensive research efforts; however, achieving both reliable efficiency and interpretability from a unified model still remains a challenging problem. Among the literature, shapelets offer interpretable and explanatory insights in the classification tasks, while most existing works ignore the differing representative power at different time slices, as well as (more importantly) the evolution pattern of shapelets. In this paper, we propose to extract time-aware shapelets by designing a two-level timing factor. Moreover, we define and construct the shapelet evolution graph, which captures how shapelets evolve over time and can be incorporated into the time series embeddings by graph embedding algorithms. To validate whether the representations obtained in this way can be applied effectively in various scenarios, we conduct experiments based on three public time series datasets, and two real-world datasets from different domains. Experimental results clearly show the improvements achieved by our approach compared with 17 state-of-the-art baselines.

cs.LG↗

arXiv2020

Time-Series Event Prediction with Evolutionary State Graph

The accurate and interpretable prediction of future events in time-series data often requires the capturing of representative patterns (or referred to as states) underpinning the observed data. To this end, most existing studies focus on the representation and recognition of states, but ignore the changing transitional relations among them. In this paper, we present evolutionary state graph, a dynamic graph structure designed to systematically represent the evolving relations (edges) among states (nodes) along time. We conduct analysis on the dynamic graphs constructed from the time-series data and show that changes on the graph structures (e.g., edges connecting certain state nodes) can inform the occurrences of events (i.e., time-series fluctuation). Inspired by this, we propose a novel graph neural network model, Evolutionary State Graph Network (EvoNet), to encode the evolutionary state graph for accurate and interpretable time-series event prediction. Specifically, Evolutionary State Graph Network models both the node-level (state-to-state) and graph-level (segment-to-segment) propagation, and captures the node-graph (state-to-segment) interactions over time. Experimental results based on five real-world datasets show that our approach not only achieves clear improvements compared with 11 baselines, but also provides more insights towards explaining the results of event predictions.

cs.LG↗

arXiv2020

Understanding Electricity-Theft Behavior via Multi-Source Data

Electricity theft, the behavior that involves users conducting illegal operations on electrical meters to avoid individual electricity bills, is a common phenomenon in the developing countries. Considering its harmfulness to both power grids and the public, several mechanized methods have been developed to automatically recognize electricity-theft behaviors. However, these methods, which mainly assess users' electricity usage records, can be insufficient due to the diversity of theft tactics and the irregularity of user behaviors. In this paper, we propose to recognize electricity-theft behavior via multi-source data. In addition to users' electricity usage records, we analyze user behaviors by means of regional factors (non-technical loss) and climatic factors (temperature) in the corresponding transformer area. By conducting analytical experiments, we unearth several interesting patterns: for instance, electricity thieves are likely to consume much more electrical power than normal users, especially under extremely high or low temperatures. Motivated by these empirical observations, we further design a novel hierarchical framework for identifying electricity thieves. Experimental results based on a real-world dataset demonstrate that our proposed model can achieve the best performance in electricity-theft detection (e.g., at least +3.0% in terms of F0.5) compared with several baselines. Last but not least, our work has been applied by the State Grid of China and used to successfully catch electricity thieves in Hangzhou with a precision of 15% (an improvement form 0% attained by several other models the company employed) during monthly on-site investigation.

cs.CY↗

arXiv2019

What Makes a Good Team? A Large-scale Study on the Effect of Team Composition in Honor of Kings

Team composition is a central factor in determining the effectiveness of a team. In this paper, we present a large-scale study on the effect of team composition on multiple measures of team effectiveness. We use a dataset from the largest multiplayer online battle arena (MOBA) game, Honor of Kings, with 96 million matches involving 100 million players. We measure team effectiveness based on team performance (whether a team is going to win), team tenacity (whether a team is going to surrender), and team rapport (whether a team uses abusive language). Our results confirm the importance of team diversity and show that diversity has varying effects on team effectiveness: although diverse teams perform well and show tenacity in adversity, they are more likely to abuse when losing than less diverse teams. Our study also contributes to the situation vs. personality debate and show that abusive players tend to choose the leading role and players do not become more abusive when taking such roles. We further demonstrate the predictive power of features based on team composition in prediction experiments.

cs.CY↗