emojinet

The code of the experiments for the EVALITA2018 challenge

Try it (semeval files should be in input/semeval_<test/train>.<labels/text> with respect to data folder):

python src/train.py \
        --workdir data \
        --max-dict 100000 \
        --max-epoch 40 \
        --semeval

Data analysis

python3 data_analysis/data_analysis.py 
        --workdir evalita_data

This script will output a file containing total number, unique number and distributions of:

• tokens
• hashtags
• mentions
• URLS
• labels

VDCNN

python3 train_vdcnn.py 
        --evalita
        --workdir evalita_data
        --max-seq-length 1024
        --pool-type k_max
        --depth 29
        --shortcut
        --bias

Results

Model	Embeddings	Accuracy	Precision	Recall	F1
Ensemble CNN subword	ft-it-our-100d	0.4375	0.3357	0.2603	0.2737
VDCNN 9	—	0.3431	0.2343	0.1750	0.1824
VDCNN 9 shortcut	—	0.3673	0.2943	0.1772	0.1926
VDCNN 9 shortcut dropout	—	0.3656	0.2997	0.1428	0.1399
BASE CNN	ft-it-our-100d	0.4435	0.4413	0.2354	0.2560
BASE CNN	provided	0.4230	0.4724	0.1905	0.2083
BASE CNN	ft-it-300d	0.4351	0.3489	0.2464	0.2673
BASE LSTM	ft-it-our-100d	0.4443	0.3666	0.2586	0.2809
BASE LSTM	ft-it-300d	0.4053	0.3167	0.2534	0.2707
BASE LSTM	provided	0.4415	0.3836	0.2408	0.2622
Most frequent user history	—	0.4396	0.4076	0.2774	0.3133
BASE LSTM User	ft-it-our-100d	0.4874	0.4343	0.3218	0.3565
BASE LSTM User (userdata)	ft-it-our-100d	0.5498	0.4872	0.4097	0.4397

Experiment log

Model	Split	Embeddings	Accuracy	Precision	Recall	F1	Remarks
BASE LSTM User	42 (def)	ft-it-our-100d	0.4875	0.4333	0.3242	0.3575	dict size: 100000
BASE LSTM User	42 (def)	ft-it-our-100d	0.4885	0.4362	0.3220	0.3571	dict size: 100000
BASE LSTM User	42 (def)	ft-it-our-100d	0.4863	0.4335	0.3193	0.3548	dict size: 100000
BASE LSTM	42 (def)	ft-it-our-100d	0.4444	0.3634	0.2626	0.2852
BASE LSTM	42 (def)	ft-it-our-100d	0.4428	0.3586	0.2638	0.2831
BASE LSTM	42 (def)	ft-it-our-100d	0.4458	0.3779	0.2494	0.2743
BASE LSTM	42 (def)	provided	0.4381	0.3662	0.2481	0.2701
BASE LSTM	42 (def)	provided	0.4456	0.3999	0.2440	0.2673
BASE LSTM	42 (def)	provided	0.4409	0.3847	0.2303	0.2492
BASE LSTM	42 (def)	ft-it-300d	0.3974	0.3102	0.2616	0.2749
BASE LSTM	42 (def)	ft-it-300d	0.4157	0.3321	0.2502	0.2721
BASE LSTM	42 (def)	ft-it-300d	0.4027	0.3077	0.2483	0.2650