Skip to content
Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

HERALD: An Annotation Efficient Method to Train User Engagement Predictors in Dialogs


This repo provides the PyTorch source code of our paper: HERALD: An Annotation Efficient Method to Train User Engagement Predictors in Dialogs (ACL 2021).

  author =  {Weixin Liang and Kai-Hui Liang and Zhou Yu},
  title =   {{HERALD:} An Annotation Efficient Method to Train User Engagement Predictors in Dialogs},
  year =    {2021},  
  booktitle = {{ACL}},
  publisher = {Association for Computational Linguistics}


Open-domain dialog systems have a user-centric goal: to provide humans with an engaging conversation experience. User engagement is one of the most important metrics for evaluating open-domain dialog systems, and could also be used as real-time feedback to benefit dialog policy learning. Existing work on detecting user disengagement typically requires hand-labeling many dialog samples. We propose HERALD, an annotation efficient framework that reframes the training data annotation process as a denoising problem. Specifically, instead of manually labeling training samples, we first use a set of labeling heuristics to automatically label training samples. We then denoise the weakly labeled data using Shapley algorithm. Finally, we use the denoised data to train a user engagement detector. Our experiments show that HERALD improves annotation efficiency significantly and achieves 86% user disengagement detection accuracy in two dialog corpora. Our implementation is available at

Stage 1: Auto-label Training Data with Heuristic Functions

Table: Our labeling heuristics designed to capture user disengagement in dialogs. A dialog turn is considereddisengaged if any of the heuristic rules applies to the user responses.

Stage 2: Denoise with Shapley Algorithm & Fine-tune


Run the following commands to create a conda environment (assuming CUDA10.1):

conda create -n herald python=3.6
conda activate herald
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
conda install matplotlib scipy
conda install -c conda-forge scikit-learn 
conda install -c conda-forge transformers
conda install pandas

Please check shapley/requirements.txt or shapley/requirements.yml for additional details about the dependencies (Note that you don't need to install all of them).

BERT-based Dialog Classifier

Please check shapley/ and shapley/ The code is built upon the github repo ABSA-PyTorch. Many thanks to the authors and developers!


python --model_name bert_spc

Running with Custom Dialog Dataset

Please check shapley/convai_data/ for supporting custom dialog dataset.

Running the Data Shapley Algorithm

Shapley algorithm computes a Shapley value for each training datum, which quantifies the contribution of each training datum to the prediction and performance of a deep network. Low Shapley value data capture outliers and corruptions. Therefore, we can identify and denoise the incorrectly-labeled data by computing their Shapley values, and then fine-tune the model on cleaned training set.

To obtain a closed-form solution of Shapley value, we extract the features of training data points and apply a K-nearest-neighbour classifier. The Shapley value of each training point can be calculated recursively as follows:

Please check shapley/ for the implementation of the shapley algorithm. Note that you need to first extract the features for training datapoints before running the K-nearest-neighbour based Shapley algorithm. In particular, the core function for calculating the single point data shapley value is:

def single_point_shapley(xt_query, y_tdev_label):
    distance1 = np.sum(np.square(X-xt_query), axis=1)
    alpha = np.argsort(distance1)
    shapley_arr = np.zeros(N)
    for i in range(N-1, -1, -1): 
        if i == N-1:
            shapley_arr[alpha[i]] = int(y[alpha[i]] == y_tdev_label) /N
            shapley_arr[alpha[i]] = shapley_arr[alpha[i+1]] + \
              ( int(y[alpha[i]]==y_tdev_label) - int(y[alpha[i+1]]==y_tdev_label) )/K * min(K,i+1)/(i+1)
    return shapley_arr

Here we use (i+1) since i starts from zero in our python implementaion.

Related Papers on Data Shapley

Beyond User Self-Reported Likert Scale Ratings: A Comparison Model for Automatic Dialog Evaluation (ACL 2020). Weixin Liang, James Zou and Zhou Yu. [PDF] [Video] [Stanford AI Lab Blog] [Slides] [Code]

Data Shapley: Equitable Data Valuation for Machine Learning. (ICML 2019). Amirata Ghorbani, James Zou. [PDF] [Video] [Poster] [Slides] [Code]


Thank you for your interest in our work! Please contact us at, for any questions, comments, or suggestions!


HERALD: An Annotation Efficient Method to Train User EngagementPredictors in Dialogs (ACL 2021)





No releases published


No packages published