File size: 2,957 Bytes
af77da5
 
e147042
 
 
 
 
 
 
 
9522772
 
 
 
 
 
 
af77da5
 
e69443a
 
af77da5
e69443a
 
 
 
 
 
9522772
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e69443a
 
 
 
 
9522772
 
e69443a
 
9522772
 
e69443a
 
9522772
 
 
 
e69443a
 
 
 
 
 
 
 
 
 
 
9522772
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
---
license: cc-by-nc-4.0
task_categories:
- question-answering
- text-generation
- summarization
language:
- en
tags:
- sports
- nba
- nfl
- reasoning
- long-context
pretty_name: SportsMetrics
size_categories:
- 1K<n<10K
---

# SportsMetrics
Benchmark data to evaluate numerical reasoning and information fusion of LLMs.

**SportsMetrics: Blending Text and Numerical Data to Understand Information Fusion in LLMs**  \
Yebowen Hu, Kaiqiang Song, Sangwoo Cho, Xiaoyang Wang, Hassan Foroosh, Dong Yu, Fei Liu   \
[*In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL'24), Bangkok, Thailand.*](https://2024.aclweb.org/program/main_conference_papers/)  \
[Arxiv Paper](https://arxiv.org/abs/2402.10979)


## Usage
```python
  from datasets import load_dataset
  
  def get_task(domain, task):
      bench_data = []
      dataset = load_dataset("huuuyeah/SportsMetrics",split="test")
      for instance in dataset:
          if instance["domain"]==domain and instance["task"]==task:
              bench_data.append(instance)
      return bench_data
  
  def message_iter(domain, task):
      bench_data = get_task(domain, task)
      if len(bench_data) == 0:
          print("No data loaded.")
          return
      
      for instance in bench_data:
          messages = [
              {"role": "system", "content": instance["system"]},
              {"role": "user", "content": instance["user"]}
          ]
          yield messages
  
      return
```

## Benchmark Tasks

The LLM is mandatorily required to generate responses in JSON format.

### Reasoning Task
- **reasoning-team_points_tracking**: (NBA) Tracking team points in one match.  
- **reasoning-key_stats_tracking**: (NBA, NFL) Tracking the key statistics for sports analytics.

### Conflicts Task
- **conflict-one_point_rule**: (NBA) All scoring actions in the competition are set to be worth only one point.  
- **conflict-swap_{num}_players**: (NBA) Swap {num} of spalyer between two teams.

### Robustness Task
- **robustness-duplicate_{prob}**: (NBA) Replicate the non-scoring move with a probability of {prob}.  
- **robustness-remove_{prob}**: (NBA) Remove the non-scoring move with a probability of {prob}.  
- **robustness-shuffled_pbp**: (NBA) Shuffle the order of all moves in play-by-play descriptions while maintain the original order of timestamps.  
- **robustness-{num}_fiction_names**: (NFL) Randomly select {num} of players from both teams and replace them with names from fiction movies. 

**Bibtex**
```
@misc{hu2024sportsmetricsblendingtextnumerical,
      title={SportsMetrics: Blending Text and Numerical Data to Understand Information Fusion in LLMs}, 
      author={Yebowen Hu and Kaiqiang Song and Sangwoo Cho and Xiaoyang Wang and Hassan Foroosh and Dong Yu and Fei Liu},
      year={2024},
      eprint={2402.10979},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2402.10979}, 
}