A Hashtag list
Following hashtags were used as distant supervision to annotate 3.8 million tweets from the Harvard Dataverse 2016 United States Presidential Election Tweet Ids Dataset . Hashtags in Appendix A.1 and A.2 were combined to label tweets with stance in favor of Hillary (or equivalently against Trump). Whereas, hashtags in Appendixs A.3 and A.4 were combined to label tweets with positive stance towards Trump (or equivalently negative stance towards Hillary).
1.1 A.1 Hashtags in favor of Hillary (+, Hillary)
hillaryforpr, imwithher2016, imwithhillary, hillaryforpresident, hillaryforamerica, hereiamwithher, estoyconella, hillarysopresidential, uniteblue, hillstorm2016, bluewave2016, welovehillary, itrusther, bluewave, hrcisournominee, itrusthillary, standwithmadampotus, momsdemandhillary, madamepresident, madampresident, imwither, herstory, republicans4hillary, hillarysoqualified, werewithher, vote4hillary, strongertogether, readyforhillary, hillafornia, unitedagainsthate, votehillary, wearewithher, republicansforhillary, hrc2016, connecttheleft, yeswekaine, voteblue2016, hillary2016, sheswithus, hillyes, iamwithher, heswithher, voteblue, hillaryaprovenleader, imwiththem, bernwithher, ohhillyes, imwithher, clintonkaine2016, whyimwithher, turnncblue, hillarystrong
1.2 A.2 Hashtags against Trump (-, Trump)
nevertrumppence, lgbthatestrumpparty, boycotttrump, orangehitler, wheresyourtaxes, poordonald, losertrump, notrumpanytime, dirtydonald, drumpf, trumpsopoor, nodonaldtrump, makedonalddrumpfagain, nastywomen, defeattrump, sleazydonald, weakdonald, unfittrump, trump20never, loserdonald, trumplies, dumbdonald, trumpliesmatter, releaseyourtaxes, crybabytrump, freethedelegates, lyingtrump, nastywomenvote, trumpleaks, stupidtrump, stoptrump, trumpthefraud, racisttrump, dumpthetrump, dumptrump, anyonebuttrump, wherertrumpstaxes, crookeddonald, treasonoustrump, antitrump, nevertrump, notrump, womentrumpdonald, nevergop, donthecon, crookeddrumpf, traitortrump, showusyourtaxes, trumptrainwreck, lyingdonald, crookedtrump, lyindonald, ripgop, trumptreason, lyintrump, chickentrump
1.3 A.3 Hashtags against Hillary (-, Hillary)
hillarysolympics, hillaryforprison, hillaryforprison2016, moretrustedthanhillary, heartlesshillary, neverclinton, handcuffhillary, queenofcorrupton, crookedhiliary, nomoreclintons, hillary4jail, fbimwithher, clintoncrimefamily, hillno, queenofcorruption, hillarylostme, ohhillno, billclintonisrapist, democratliesmatter, lyingcrookedhillary, hypocritehillary, crookedclintons, hillarylies, neverhilary, shelies, releasethetranscripts, stophillary2016, riskyhillary, hillaryliedpeopledied, corrupthillary, hillary4prison2016, nohillary2016, wehatehillary, whatmakeshillaryshortcircuit, crookedhillaryclinton, deletehillary, dropouthillary, lyinhillary, hillaryliesmatter, nevereverhillary, stophillary, neverhilllary, clintoncorruption, clintoncrime, notwithher, hillary2jail, imnotwithher, lockherup, corruptclintons, indicthillary, sickhillary, crookedhilary, crookedhillary, hillaryrottenclinton, theclintoncontamination, lyinghillary, clintoncollapse, clintoncrimefoundation, neverhillary, criminalhillary, crookedclinton, hillary4prison, killary, iwillneverstandwithher
1.4 A.4 Hashtags in favor of Trump (+, Trump)
trumppence2016, trumpstrong, donaldtrumpforpresident, rednationrising, deplorablesfortrump, makeamericaworkagain, latinos4trump, trumpiswithyou, blacks4trump, feelthetrump, votetrumppence2016, bikers4trump, votetrump2016, votetrumppence, americafirst, trumpcares, draintheswamp, votetrumpusa, trumppence16, gaysfortrump, buildthewall, trump2016, trumpwon, alwaystrump, onlytrump, maga3x, veteransfortrump, latinosfortrump, cafortrump, gays4trump, makeamericasafeagain, latinoswithtrump, trump16, woman4trump, womenfortrump, makeamericagreat, votegop, makeamericagreatagain, maga, trumptrain, gotrump, bikersfortrump, votetrumppence16, feminineamerica4trump, trumpwins, imwithhim, buildthatwall, babesfortrump, america1st, securetheborder, vets4trump, democrats4trump, women4trump, trumpforpresident, magax3, blacksfortrump, heswithus, presidenttrump, votetrump
B Evaluation on SemEval2016 Task 6 test-set
We evaluate our best performing CNN model on SemEval2016 Task 6 test-set with target ’Hillary Clinton’. This dataset contains 295 tweets with gold labels of ’AGAINST’, ’FAVOR’ or ’NEUTRAL’ as stance towards Hillary. Since our CNN is trained on ’FAVOR’ and ’AGAINST’ stance, following the same experimental setup, we extract tweets that are in favor ( or against the target (hillary clinton). We find that our model model performs well on the test-set with weighted F1-score of 0.75 (See Table 3 for the confusion matrix).
C Topic list
money, japan, justice, climate change, economy, rapist, healthcare, podesta, obamacare, abortion, foreign, women, nato, cyber, russia, weapon, podesta email, voter fraud, benghazi, iran, assault, email, blm, gun, tape, job, podestaemail, middle east, police, p2, climatechange, 2ndamendment, amendment, audit, lgbt, 2nd amendment, appoint, climate, nafta, war, second amendment, black, middle class, mosul, tax, nuke, 2a, scotus, korea, isis, iraq, haiti, putin, trade, paytoplay, voterfraud, woman, china, law, nuclear, syria, secondamendment, rig, debt