SE-PQA: a Resource for Personalized Community Question Answering

Personalization in Information Retrieval is a topic studied for a long time. Nevertheless, there is still a lack of high-quality, real-world datasets to conduct large-scale experiments and evaluate models for personalized search. This paper contributes to fill this gap by introducing SE-PQA (StackExchange - Personalized Question Answering), a new resource to design and evaluate personalized models related to the two tasks of community Question Answering (cQA). The contributed dataset includes more than  1 million queries and 2 million answers,  annotated with a rich set of features modeling the social interactions among the users of a popular cQA platform. We describe the characteristics of SE-PQA and detail the features associated with both questions and answers. We also provide reproducible baseline methods for the cQA task based on the resource, including deep learning models and personalization approaches. The results of the preliminary experiments conducted show the appropriateness of SE-PQA to train effective cQA models; they also show that personalization improves remarkably the effectiveness of all the methods tested. Furthermore, we show the benefits in terms of robustness and generalization of combining data from multiple communities for personalization purposes.

Performance on all communities separately:

Community
Model (BM25 +)
P@1
NDCG@3
NDCG@10
R@100
MAP@100
$\lambda$

Academia
MiniLM
0.438
0.382
0.395
0.489
0.344
(.1,.9)

MiniLM + TAG
0.453
0.392
0.403
0.489
0.352
(.1,.8,.1)

Anime
MiniLM + TAG
0.650
0.682
0.714
0.856
0.683
(.1,.9,.0)

Apple
MiniLM
0.327
0.351
0.381
0.514
0.349
(.1,.9)

MiniLM + TAG
0.335
0.361
0.389
0.514
0.357
(.1,.8,.1)

Bicycles
MiniLM
0.405
0.380
0.421
0.600
0.365
(.1,.9)

MiniLM + TAG
0.436
0.405
0.441
0.600
0.386
(.1,.8,.1)

Boardgames
MiniLM
0.681
0.694
0.728
0.866
0.692
(.1,.9)

MiniLM + TAG
0.696
0.702
0.736
0.866
0.699
(.1,.8,.1)

Buddhism
MiniLM + TAG
0.490
0.387
0.397
0.544
0.334
(.3,.7,.0)

Christianity
MiniLM
0.534
0.505
0.555
0.783
0.497
(.2,.8)

MiniLM + TAG
0.549
0.521
0.564
0.783
0.507
(.1,.8,.1)

Cooking
MiniLM
0.600
0.567
0.600
0.719
0.553
(.1,.9)

MiniLM + TAG
0.619
0.583
0.614
0.719
0.568
(.1,.8,.1)

DIY
MiniLM
0.323
0.313
0.346
0.501
0.302
(.1,.9)

MiniLM + TAG
0.335
0.324
0.356
0.501
0.312
(.1,.8,.1)

Expatriates
MiniLM + TAG
0.596
0.653
0.682
0.832
0.645
(.1,.9,.0)

Fitness
MiniLM + TAG
0.568
0.575
0.613
0.760
0.567
(.2,.8,.0)

Freelancing
MiniLM + TAG
0.513
0.472
0.506
0.654
0.457
(.1,.9,.0)

Gaming
MiniLM
0.510
0.534
0.562
0.686
0.532
(.1,.9)

MiniLM + TAG
0.519
0.547
0.571
0.686
0.541
(.1,.8,.1)

Gardening
MiniLM
0.344
0.362
0.396
0.520
0.359
(.1,.9)

MiniLM + TAG
0.345
0.369
0.399
0.520
0.363
(.1,.8,.1)

Genealogy
MiniLM + TAG
0.592
0.605
0.631
0.779
0.594
(.3,.7,.0)

Health
MiniLM + TAG
0.718
0.765
0.797
0.934
0.765
(.2,.8,.0)

Gaming
MiniLM
0.510
0.534
0.562
0.686
0.532
(.1,.9)

MiniLM + TAG
0.519
0.547
0.571
0.686
0.541
(.1,.8,.1)

Hermeneutics
MiniLM
0.589
0.538
0.593
0.828
0.526
(.2,.8)

MiniLM + TAG
0.632
0.570
0.617
0.828
0.552
(.1,.8,.1)

Hinduism
MiniLM
0.388
0.415
0.459
0.686
0.416
(.2,.8)

MiniLM + TAG
0.382
0.410
0.457
0.686
0.412
(.1,.8,.1)

History
MiniLM + TAG
0.740
0.735
0.764
0.862
0.730
(.2,.8,.0)

Hsm
MiniLM + TAG
0.666
0.707
0.737
0.870
0.690
(.2,.8,.0)

Interpersonal
MiniLM + TAG
0.663
0.617
0.653
0.739
0.604
(.2,.8,.0)

Islam
MiniLM
0.382
0.412
0.453
0.642
0.410
(.1,.9)

MiniLM + TAG
0.395
0.427
0.464
0.642
0.421
(.1,.8,.1)

Judaism
MiniLM + TAG
0.363
0.387
0.432
0.649
0.388
(.2,.8,.0)

Law
MiniLM
0.663
0.647
0.678
0.803
0.639
(.2,.8)

MiniLM + TAG
0.677
0.657
0.687
0.803
0.649
(.1,.8,.1)

Lifehacks
MiniLM
0.714
0.601
0.617
0.703
0.553
(.1,.9)

MiniLM + TAG
0.714
0.621
0.631
0.703
0.568
(.1,.8,.1)

Linguistics
MiniLM + TAG
0.584
0.588
0.630
0.794
0.587
(.2,.8,.0)

Literature
MiniLM + TAG
0.871
0.878
0.889
0.934
0.876
(.3,.7,.0)

Martialarts
MiniLM
0.630
0.599
0.645
0.796
0.596
(.1,.9)

MiniLM + TAG
0.640
0.628
0.660
0.796
0.612
(.1,.8,.1)

Money
MiniLM
0.545
0.535
0.563
0.706
0.515
(.2,.8)

MiniLM + TAG
0.559
0.542
0.571
0.706
0.523
(.1,.8,.1)

Movies
MiniLM
0.713
0.722
0.753
0.865
0.724
(.1,.9)

MiniLM + TAG
0.728
0.735
0.762
0.865
0.735
(.1,.8,.1)

Music
MiniLM
0.508
0.447
0.476
0.602
0.418
(.2,.8)

MiniLM + TAG
0.522
0.460
0.486
0.602
0.427
(.1,.8,.1)

Musicfans
MiniLM + TAG
0.531
0.531
0.560
0.693
0.539
(.1,.9,.0)

Opensource
MiniLM
0.574
0.593
0.621
0.771
0.581
(.2,.8)

MiniLM + TAG
0.577
0.598
0.622
0.771
0.581
(.1,.8,.1)

Outdoors
MiniLM + TAG
0.681
0.643
0.675
0.819
0.629
(.1,.9,.0)

Parenting
MiniLM + TAG
0.485
0.430
0.452
0.602
0.399
(.1,.9,.0)

Pets
MiniLM
0.509
0.531
0.565
0.685
0.523
(.1,.9)

MiniLM + TAG
0.519
0.549
0.581
0.685
0.541
(.1,.8,.1)

Philosophy
MiniLM + TAG
0.568
0.514
0.546
0.707
0.491
(.2,.8,.0)

Politics
MiniLM + TAG
0.659
0.630
0.659
0.814
0.608
(.1,.9,.0)

Rpg
MiniLM
0.657
0.646
0.685
0.849
0.640
(.2,.8)

MiniLM + TAG
0.677
0.660
0.695
0.849
0.651
(.1,.8,.1)

Scifi
MiniLM
0.532
0.563
0.596
0.745
0.559
(.2,.8)

MiniLM + TAG
0.549
0.574
0.606
0.745
0.569
(.1,.8,.1)

Skeptics
MiniLM + TAG
0.862
0.869
0.887
0.969
0.867
(.2,.8,.0)

Sound
MiniLM
0.377
0.410
0.451
0.626
0.405
(.2,.8)

MiniLM + TAG
0.380
0.423
0.454
0.626
0.413
(.1,.8,.1)

Sports
MiniLM
0.673
0.721
0.756
0.902
0.724
(.2,.8)

MiniLM + TAG
0.692
0.743
0.775
0.902
0.740
(.1,.8,.1)

Sustainability
MiniLM
0.657
0.677
0.735
0.895
0.675
(.1,.9)

MiniLM + TAG
694
0.716
0.763
0.895
0.706
(.1,.8,.1)

Travel
MiniLM + TAG
0.546
0.547
0.576
0.700
0.530
(.1,.9,.0)

Vegetarianism
MiniLM + TAG
0.623
0.626
0.678
0.869
0.641
(.3,.7,.0)

Woodworking
MiniLM + TAG
0.656
0.654
0.692
0.847
0.645
(.2,.8,.0)

Workplace
MiniLM + TAG
0.574
0.444
0.429
0.495
0.359
(.3,.7,.0)

Writers
MiniLM + TAG
0.561
0.490
0.516
0.644
0.466
(.2,.8,.0)

Average
MiniLM
0.519
0.506
0.536
0.677
0.492
-

MiniLM + TAG
0.530
0.515
0.544
0.677
0.500
-