You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Automating Alternative Generation in Decision-Making

This repo contains data for alternative generation in decision making from the paper:

@inproceedings{
kostiukemnlp2025,
title={Automating Alternative Generation in Decision-Making},
author={Yevhen Kostiuk, Clara Seyfried, Chris Reed},
booktitle={EMNLP 2025},
year={2025}
}

To avoid dataset leak and ensure the fair benchmarking of the newer LLMs, the dataset is available once you are submit a form. The form is automatically provides you access.

Loading Data

The original posts and comments are accessible from ConvoKit, specifically Advice subreddit.

To match the data, use id in utterences field with any column containing id in our corpora.

The dataset contains 3 configs: posts_meta, replies, solutions, and solutions_scores.

posts_meta

posts_meta dataset of reddit post contents, titles, updates from the post's author (if any), and post_ids.

from datasets import load_dataset

ds = load_dataset(hf_link, "posts_meta")

df = ds['train'].to_pandas()

The final df looks like following (first two rows):

    title	body	update	post_id
0	Should I drop out of high school?	Here's the deal. I'm fluent in Java, JavaScript, and Python, am an artist, a songwriter, and a voice actor. I know I must sound like a total asshole, but I can't stand high school. The way I see it, if I drop out and get my GED and then spend the next 3 years of my life (sophomore year, junior year, and senior year) sharpening my skills and building a resume of art, music, and video games, then by the time I'm 18, I will be able to find a job. Even if I cant, I can be a web designer or some other independent type deal. I just feel so out of place in high school. What do you guys think?		29xrdg
1	I got kicked out of my home at 17, what do I do now?	"Let me give some background to explain my situation. I'm 17 (I turn 18 in 6 months) and live in Michigan, but I've graduated highschool and currently am enrolled in a local college as a sophmore. All my savings have gone towards tuition (paying out of my pocket, my parents dont contribute). As of right now I have about $100 in my bank account, the clothes on my back, and my laptop/phone.

Over the past two years I've been clashing with my dad a lot, most of the time it's just over small things but it happens very frequently. Today was finally the straw that broke the camels back, and a very small incident turned into a huge argument and he finally kicked me out of the house. I've never seen him that mad before. I'm honestly quite at a loss at what to do, I'm just going to my classes and sleeping in some remote corner on campus for today. Realistically there are facilities I have access to as a student where I can shower every morning, and there are couches on campus I can sleep on, but food/clothes will become a problem. I don't have any other relatives that live nearby and I was never super social so I don't have any friends good enough that I would feel comfortable asking them for a place to sleep for a little bit until I figure things out.

I have a well paying co-op opportunity lined up in a year with a company, but I don't have a job currently. However, this co-op requires me to maintain enrollment (and good grades) in my university. Although things are payed for this semester, coming up with the money for next semester (and money to live off of until then) is what I'm worried about. Then there's all the legal stuff with social security, taxes, life insurance etc, which I have no clue about. I'm not even sure if this is the right place to post. Reddit, I need some help. What do I do now?"	Thank you everyone for your advice and offers to help! I decided to follow the majority of advice and research the laws in Michigan, and you're right, my dad can't legally kick me out unless I'm emancipated or 18, and not without a notice to vacate. Since I don't want to completely ruin our relationship forever I won't go to police, child services, etc without talking to him first. I decided to wait a day for both of us to cool down a bit and talk to him tomorrow. I would firstly apologize, but will bring up all the legal stuff if necessary. If he still refuses at that point, I would go to the police. Thanks again everyone, I was very worried about what would happen now, and seeing all this really helped.	2tyzyr

replies

replies dataset contains interaction of the post author with the commentor. It contains:

  • solution_text: standartized solution text among comments for the problem; If none provided - NO_SOLUTION;
  • alternative_num: number of alternatives proposed in the text. For example, if commentor suggests buy carrot or buy salad, there will be two solutions: buy carrot with alternative_num 1, and buy salad with alternative_num 2. Sometimes, they are nested, could include alternatives in alternatives. In that case, we use letters: 1a, 1b etc.
  • response_id: id of the comment, where the solution was proposed.
  • author_answer_id: id of the author response to the comment (if any).
  • author_did: binary variable, if author indicated a commitment to do what was proposed in the comment solution. 1 if yes, 0 if no.
  • is_update: binary variable, indicates if the comment is an update from the author's original post. In this case, esponse_id and author_answer_id are set to OP (original post).
  • post_id: id of the post.
from datasets import load_dataset

ds = load_dataset(hf_link, "posts_meta")

df = ds['train'].to_pandas()

The final df looks like following (first two rows):

    solution_text	alternative_num	response_id	author_answer_id	author_did	is_update	post_id	suggested_comment_id	__index_level_0__
0	Get a GED if you are too smart for school.	-1	cipim14	cipiniq	0	0	29xrdg	cipim14	9
1	NO_SOLUTION	-1	cipj2vq	cipj3l6	0	0	29xrdg		10

solutions

solutions dataset contains labeled comments with the solutions people proposed. It contains:

  • solution_text: standartized solution text among comments for the problem; If none provided - NO_SOLUTION;
  • alternative_num: number of alternatives proposed in the text. For example, if commentor suggests buy carrot or buy salad, there will be two solutions: buy carrot with alternative_num 1, and buy salad with alternative_num 2. Sometimes, they are nested, could include alternatives in alternatives. In that case, we use letters: 1a, 1b etc.
  • post_id: id of the post.
  • comment_id: id of the comment, where solution was proposed.
  • annotator: annotator id (0 or 1).
from datasets import load_dataset

ds = load_dataset(hf_link, "posts_meta")

df = ds['train'].to_pandas()

The final df looks like following (first two rows):

    solution_text	alternative_num	comment_id	post_id	annotator	__index_level_0__
0	Talk to your parents during during the time when you are all together and attentive and tell them that you feel very grateful for the opportunity they have given you, you want to continue living here and respect the rules.	-1	dv2l133	81ef3x	0	121
1	When the spanking instance starts, stand up, look your parents in the eye and say 'no, this isn't happening again'.	-1	dv2ow3p	81ef3x	0	122

scores

solutions_scores dataset contains calcaulted per-solution scores required for the score-based matric. The scores are based on the upvote rates. It contains:

  • solution_texts: list of str, standartized solution text among comments for the problem;
  • raw_sum_scores: list of ints, number of upvotes for each solution text (sum of upvotes of all comments that contained this solution).
  • post_id: id of the post.
  • softmax_scores: softmax applied on raw_sum_scores.
  • norm_scores: normalization applied on raw_sum_scores.
from datasets import load_dataset

ds = load_dataset(hf_link, "posts_meta")

df = ds['train'].to_pandas()

The final df looks like following (first two rows):

    solution_texts	raw_sum_scores	softmax_scores	norm_scores	post_id
0	"['You could stay out of this situation.'
 'You could send your concerns to the superintendent.'
 'ask for your money back as it was not spent on the excursion.'
 'Teach your kid about the dangers of soda and overall safety'
 'send your concerns to the superintendent, but be direct and not so emotional.Outline your concerns why they are health issues got these kids'
 'You could discuss your concerns with the school in a friendly manner.'
 'You could push for a petition to amend the county law to extend the law to include field trips and extracurricular activities.'
 'You could file a complaint with the state or city.'
 ""If your school has a website, take some time to explore it and see if there is a number you can call or a document you can borrow and make copies of.  Ask people other than the ones that you've already spoken to for official documents.""
 ""You could push for a petition to amend the county law to extend the law to include field trips and extracurricular activities. Once you have your petition, you'll need to round up the other parents and present it to them. Just state the facts, that you are worried about the lack of parental control over what is happening to your children.""
 'Go up the hierarchy until the solution will be found: principal, superintendents office, state dept of education, federal dept of educationm state department of agriculture,  USDA, federal dept of agriculture'
 'Complain to the school about the situation'
 'You could withdraw your child from the school and teach them yourself.']"	[42.  6.  1.  6.  6.  6.  7.  1.  1.  1.  2.  1.  2.]	"[1.00000000e+00 2.31952270e-16 1.56288218e-18 2.31952270e-16
 2.31952270e-16 2.31952270e-16 6.30511685e-16 1.56288218e-18
 1.56288218e-18 1.56288218e-18 4.24835413e-18 1.56288218e-18
 4.24835413e-18]"	"[0.51219512 0.07317073 0.01219512 0.07317073 0.07317073 0.07317073
 0.08536585 0.01219512 0.01219512 0.01219512 0.02439024 0.01219512
 0.02439024]"	0
1	"['Get a GED if you are too smart for school.'
 'Do high school, and spend all your free time teaching yourself and sharpening your skills'
 'do not drop out of high school'
 'drop the Advanced Placement (AP) classes (except computer science) then use the free time to build a resume.'
 ""try really hard to convince my parents to let me switch schools, a school that doesn't give me anxiety.""
 'Drop the course, study the material in your free time, and take it next year.'
 'drop out of high school'
 'Try to take some college courses through your high school'
 'look into the early graduation']"	[ 0.  2. 25.  3.  3.  1.  3.  1.  1.]	"[1.38879437e-11 1.02618795e-10 1.00000000e+00 2.78946810e-10
 2.78946810e-10 3.77513437e-11 2.78946810e-10 3.77513437e-11
 3.77513437e-11]"	"[0.         0.05128205 0.64102564 0.07692308 0.07692308 0.02564103
 0.07692308 0.02564103 0.02564103]"	1

The scores were calcaulted via following script:

import torch
from datasets import load_dataset

ds = load_dataset(hf_link, "posts_meta")

final_comment_df = ds['train'].to_pandas()


posts_solutions_upvotes_norms = {}

with torch.no_grad():
    for post_id, post_df_ in final_comment_df.groupby("post_id"):
        posts_solutions_upvotes_norms[post_id] = {
            "texts": [],
            "raw_sum_scores": []
        }
        
        for solution_text in post_df_["solution_text"].unique():
            if solution_text != "NO_SOLUTION":
                solution_sum_score = 0
                solution_comments = post_df_[post_df_["solution_text"] == solution_text]['comment_id'].unique()
                for comm_id in solution_comments:
                    if comm_id.startswith("OP"):
                        solution_sum_score += post_comments_upvotes[post_id]["OP"]
                    else:
                        solution_sum_score += post_comments_upvotes[post_id][comm_id]
                posts_solutions_upvotes_norms[post_id]["texts"].append(solution_text)
                posts_solutions_upvotes_norms[post_id]["raw_sum_scores"].append(float(solution_sum_score))
        
        posts_solutions_upvotes_norms[post_id]["softmax_scores"] = torch.nn.functional.softmax(
            torch.Tensor(posts_solutions_upvotes_norms[post_id]["raw_sum_scores"]).reshape(1, -1)
        ).numpy().reshape(-1).tolist()

        posts_solutions_upvotes_norms[post_id]["norm_scores"] = [
            x/sum(posts_solutions_upvotes_norms[post_id]["raw_sum_scores"]) for x in posts_solutions_upvotes_norms[post_id]["raw_sum_scores"]
        ]
Downloads last month
5