Real-world instructions with multiple constraints pose a significant challenge to existing large language models (LLMs). An observation is that the LLMs exhibit dramatic performance fluctuation when disturbing the order of the incorporated constraints. Yet, none of the existing works has systematically investigated this position bias problem in the field of multi-constraint instruction following. To bridge this gap, we design a probing task where we quantitatively measure the difficulty distribution of the constraints by a novel Difficulty Distribution Index (CDDI). Through the experimental results, we find that LLMs are more performant when presented with the constraints in a “hard-to-easy” order. This preference can be generalized to LLMs with different architecture or different sizes of parameters. Additionally, we conduct an explanation study, providing an intuitive insight into the correlation between the LLM’s attention and constraint orders. Our code and dataset are publicly available at https://github.com/meowpass/PBIF.

It is crucial for large language models (LLMs) to follow instructions that involve multiple constraints. In real-world scenarios, user instructions often contain soft constraints, which are semantically related and cannot be rule-based verified, posing challenges for LLMs. To enhance the soft constraint following ability of LLMs, we initially design a pipeline to construct datasets with high-quality outputs for instructions containing soft constraints automatically. Additionally, to fully utilize the positive and negative samples generated during the data construction process, we choose Direct Preference Optimization (DPO) as the training method. Furthermore, taking into account the difficulty of soft constraints indicated by the number of constraints, we design a curriculum learning training paradigm based on the constraint quantity. We experimentally evaluate the effectiveness of our methods in improving LLMs’ soft constraint following ability and analyze the factors driving the improvements.

The Human Value Detection shared task\cite{kiesel:2023} aims to classify whether or not the argument draws on a set of 20 value categories, given a textual argument. This is a difficult task as the discrimination of human values behind arguments is often implicit. Moreover, the number of label categories can be up to 20 and the distribution of data is highly imbalanced. To address these issues, we employ a multi-label classification model and utilize a class-balanced loss function. Our system wins 5 first places, 2 second places, and 6 third places out of 20 categories of the Human Value Detection shared task, and our overall average score of 0.54 also places third. The code is publicly available at \url{https://www.github.com/diqiuzhuanzhuan/semeval2023}.