Topic Extraction from Steem Tags

Topic Extraction from Steem Tags

Or: Dimensionality Reduction on the Blockchain

Or: A Not-So-Gentle Introduction to Non-Negative Matrix Factorization

Yesterday, I used covariance to analyze which Steem tags were used together. Today, I'll introduce a different technique and explore the related keywords that it exposes.

NNMF

Non-Negative Matrix Factorization is a dimensionality reduction technique; that is, it compresses data to a smaller number of features. When this process works well, it preserves the original data set, by explaining them as combinations of "hidden" features extracted from the data. All the features and combinations are given positive weights, which can make them easier to interpret compared to a technique like Singular Value Decomposition, which produces both positive and negative weights.

In linear algebra terms, NNMF is just matrix multiplication: V = WH.

V is a m x n matrix that represents the input data (or, an approximation of it.) Typically the rows are observations or data points and the columns are features that have been measured. In my experiment, m = 200000 discussions and n = 5180 common tags. A '1' entry in the matrix means that a particular discussion had that topic specified; otherwise it is '0'. (I had to cut down from my original data set of 1.1 million discussions due to lack of memory on the VM where I performed the analysis.)

W is a m x p matrix that maps each observation to a weighted sum of the "p" computed hidden features or "topics". In practice, a data scientist has to zero in on a good value for p; I choose p = 100 arbitrarily. So, this matrix maps from Steem conversation to the dimensionally-reduced topic set. We know that users use tags that are correlated, so if NNMF does a good job we should find tags that are semantically similar appearing in the same topic.

H is a p x n matrix that maps each topic back to the original features, the tags. You can think of it as an index saying what each topic "means".

The original matrix had about a billion entries though my implementation stores them sparsely: 200000 * 5180 = 1,036,000,000. The NNMF decomposition tries to store the same amount of data in 200000 * 100 + 100 * 5180 = 20,518,000 entries. If all the original measurements were real numbers, this would be a lot less data! But the original data is actually just binary, and the NNMF matrices are filled with floating-point numbers.

The scikit-learn NNMF implementation

I used the Python package scikit-learn which provides an NNMF implementation which supports a couple different algorithms. It takes the matrix V as input and finds the best approximation of V in the decomposed form WH. After the input matrix is set up, it's just a few lines to run the algorithm and extract the desired matrices:

model = NMF( n_components = 100, init = 'random' )
W = model.fit_transform( V )
H = model.components_

>>> V.shape
(200000, 5180)
>>> W.shape
(200000, 100)
>>> H.shape
(100, 5180)
>>> model.reconstruction_err_
532.8368170706713

The reconstruction error says how close to the original V we got using the "Frobenius Norm", which is a generalization of Euclidean distance. We could try different values of p (n_components) to try to find a sweet spot where the number of topics is still small but the error is improved.

Steem topics from tags

The resulting output is quite noisy since the algorithm tries to optimize by making small changes to the weights; most topics have a lot of tags associated with them with very small values. I picked an arbitrary threshold of 0.1 for the data shown below and excluded weights lower than that.

Many weights are greater than 1.0, even though the original data has only 0 and 1! But the W matrix can and does use fractional values, not just to mix topics but also select a single topic at the appropriate weight. The number itself isn't as important as the relative weight within a topic.

Language-based topics

The algorithm did a good job finding tags related to a particular language or nationality and grouping them into topics:

topic number	highest-weight tag	other tags
29	kr-newbie, 3.017	kr-writing, 0.344 ; kr-life, 0.334 ; kr, 0.125 ; kr-daily, 0.119
66	kr, 7.209	coinkorea, 0.766 ; kr-life, 0.548 ; kr-writing, 0.483 ; kr-daily, 0.340 ; kr-event, 0.301 ; kr-art, 0.240 ; kr-pen, 0.236 ; kr-coin, 0.228 ; kr-diary, 0.157 ; tooza, 0.135 ; muksteem, 0.132 ; kr-news, 0.113 ; kr-travel, 0.113 ; kr-gazua, 0.112
84	jjangjjangman, 5.574	muksteem, 0.517 ; kr-life, 0.419 ; kr-event, 0.308 ; kr-gazua, 0.186 ; kr-funfun, 0.184 ; dev, 0.156 ; it-news, 0.154 ; kr-travel, 0.144 ; kr-diary, 0.136 ; sharehows, 0.126 ; kr-overseas, 0.122 ; kr-youth, 0.121 ; kr-series, 0.109 ; kr-art, 0.107 ; kr-food, 0.102 ; kr-easy, 0.101
46	cn, 4.038	cn-reader, 0.877 ; lizhi, 0.171 ; teammalaysia, 0.155 ; cn-book, 0.155 ; cn-malaysia, 0.153 ; book, 0.136 ; stats, 0.123
79	deutsch, 4.702	steemit-austria, 0.557 ; german, 0.161 ; politik, 0.113 ; austria, 0.100
86	spanish, 5.171	venezuela, 0.284 ; steemitvenezuela, 0.120 ; talentclub, 0.114
59	castellano, 4.063	venezuela, 1.064 ; literatura, 0.178 ; provenezuela, 0.114 ; proconocimiento, 0.105

Single-Tag topics

Tags such as #blog, #steemit, #photo, #phography, #life, are all common enough that the dimensionality reduction created features just for them, without any other significant tags.

Community interests

Photography appears in a couple different topics and NMF did a reasonable job associating semantically similar tags together:

topic number	highest-weight tag	other tags
1	photography, 2.670
38	photo, 4.016
19	colourfulphotography, 4.443	bescouted, 0.745 ; thealliance, 0.550 ; portraitphotography, 0.458 ; yourluckyphotos, 0.451 ; streetphotography, 0.430 ; sevendaybnwchallenge, 0.412 ; venezuela, 0.325 ; photocontests, 0.163 ; photoworld, 0.149 ; ru, 0.146 ; vn, 0.140 ; ua, 0.135
25	macrophotography, 6.246	acehmacro, 0.649 ; photocircle, 0.343 ; macro, 0.264 ; yourluckyphotos, 0.198 ; photocontests, 0.144
35	animals, 7.821	animalphotography, 1.388 ; cats, 0.852 ; pets, 0.672 ; dogs, 0.486 ; cat, 0.397 ; dog, 0.370 ; dailypetphotography, 0.319 ; homesteading, 0.311 ; cute, 0.288 ; birds, 0.288 ; wildlife, 0.154 ; pet, 0.136 ; caturday, 0.134 ; monomad, 0.102
45	landscapephotography, 6.021	photocircle, 0.470 ; landscape, 0.364 ; yourluckyphotos, 0.225 ; photocontest, 0.176 ; newbieresteemday, 0.127 ; naturephotography, 0.126 ; mountains, 0.119 ; photofriend, 0.110 ; photomatic, 0.107 ; sunset, 0.106
73	animal, 5.792	animalphotography, 0.563 ; cat, 0.347 ; dog, 0.161 ; acehmacro, 0.111
21	beauty, 4.672	girl, 0.441 ; woman, 0.350 ; japanese, 0.258 ; fashion, 0.233 ; fitness, 0.186 ; model, 0.160 ; portraitphotography, 0.115 ; makeup, 0.115 ; bescouted, 0.110 ; photocontest, 0.102
53	smartphonephotography, 6.191	yourluckyphotos, 0.130 ; smartphotography, 0.110

Porn terms cluster together strongly (suggesting perhaps a lack of originality in tag selection), with some overlap with #hot:

topic number	highest-weight tag	other tags
98	nsfw, 2.889	sexy, 1.681 ; porn, 1.002 ; sex, 0.678 ; nude, 0.497 ; girl, 0.420 ; girls, 0.311 ; hot, 0.276 ; adult, 0.274 ; woman, 0.250 ; dporn, 0.205 ; hentai, 0.202 ; japanese, 0.182 ; hardcore, 0.172 ; model, 0.165 ; anime, 0.157 ; pussy, 0.155 ; boobs, 0.138 ; asmr, 0.137 ; erotica, 0.117 ; manga, 0.106 ; ass, 0.105
5	hot, 4.229	trending, 3.951 ; promoted, 2.900 ; girls, 0.181 ; sexy, 0.176 ; asmr, 0.165 ; viral, 0.123

Science-related topics get grouped together too in a couple overlapping topics:

topic number	highest-weight tag	other tags
33	technology, 3.760	science, 0.351 ; business, 0.155 ; steemstem, 0.120 ; steemdunk, 0.114
78	science, 3.635	education, 3.413 ; steemstem, 0.751 ; steemiteducation, 0.622 ; people, 0.607 ; space, 0.224 ; sndbox, 0.159 ; psychology, 0.159 ; physics, 0.137 ; philosophy, 0.119 ; stemng, 0.118

The complete output is included below.

Conclusion

Non-negative Matrix Factorization is a standard dimensionality-reduction technique which I have demonstrated on Steem tags and discussions. It is frequently used for this sort of semantic clustering; in fact, that's the example scikit-learn provides! The decomposed form of the original data can be used for other purposes as well, such as similarity detection or recommendations; I am thinking of analyzing upvotes according to these topics rather than the raw tags. A more ambitious project would be to apply NNMF to word frequencies or occurence within the discussions themselves, not just the attached tags.

Complete topic list

topic number	highest-weight tag	other tags
0	life, 1.952
1	photography, 2.670
2	artzone, 1.136	slothicorn, 0.110
3	untalented, 0.733	steemitfamilyph, 0.138
4	whalepower, 1.033
5	ocd-resteem, 2.955	promo-steem, 0.375 ; slothicorn, 0.283 ; homesteading, 0.215 ; ita, 0.173 ; adsactly, 0.149 ; gardening, 0.102
6	blockchain, 2.670
7	love, 2.222
8	bitcoin, 3.287	ethereum, 0.207
9	introduceyourself, 1.294	introducemyself, 0.235
10	beautiful, 2.843	quotes, 0.735 ; amazing, 0.449 ; words, 0.349 ; minds, 0.334 ; lovefriday, 0.263 ; japanese, 0.249 ; happy, 0.232 ; girl, 0.210 ; woman, 0.155 ; nice, 0.107
11	art, 3.709
12	food, 3.137	recipe, 0.308 ; foodphotography, 0.253 ; cooking, 0.174
13	travel, 2.515
14	nature, 5.634
15	dlive, 1.631	dlive-video, 1.560 ; learning, 0.317 ; , 0.172
16	dtube, 3.332	vlog, 0.277 ; dtubedaily, 0.275 ; onelovedtube, 0.252 ; steempowerment, 0.117
17	flower, 2.820
18	entertainment, 1.933	people, 0.364 ; movies, 0.106
19	colourfulphotography, 4.443	bescouted, 0.745 ; thealliance, 0.550 ; portraitphotography, 0.458 ; yourluckyphotos, 0.451 ; streetphotography, 0.430 ; sevendaybnwchallenge, 0.412 ; venezuela, 0.325 ; photocontests, 0.163 ; photoworld, 0.149 ; ru, 0.146 ; vn, 0.140 ; ua, 0.135
20	meme, 2.519	dmania, 1.330 ; memes, 0.139 ; funny, 0.134 ; memeitlol, 0.115
21	beauty, 4.672	girl, 0.441 ; woman, 0.350 ; japanese, 0.258 ; fashion, 0.233 ; fitness, 0.186 ; model, 0.160 ; portraitphotography, 0.115 ; makeup, 0.115 ; bescouted, 0.110 ; photocontest, 0.102
22	music, 5.211	comedy, 0.411 ; geme, 0.405 ; openmic, 0.291 ; guitar, 0.159 ; song, 0.157 ; rock, 0.146 ; dance, 0.109 ; hiphop, 0.108 ; musicdiscovery, 0.106
23	esteemapp, 3.557	information, 0.517 ; religious, 0.456 ; terp, 0.138 ; kmk, 0.102
24	crypto, 4.410	cryptocurrency, 4.403 ; crypto-news, 0.433
25	macrophotography, 6.246	acehmacro, 0.649 ; photocircle, 0.343 ; macro, 0.264 ; yourluckyphotos, 0.198 ; photocontests, 0.144
26	drawing, 3.151	painting, 0.725 ; sketch, 0.335 ; creativity, 0.270 ; slothicorn, 0.153 ; anime, 0.131 ; illustration, 0.111 ; coloredpencil, 0.107
27	health, 4.736	fitness, 0.300 ; tips, 0.103
28	busy, 6.683
29	kr-newbie, 3.017	kr-writing, 0.344 ; kr-life, 0.334 ; kr, 0.125 ; kr-daily, 0.119
30	cervantes, 3.487
31	steem, 7.676	steempress, 1.107 ; sbd, 0.151
32	contest, 6.422	challenge, 0.364 ; photobombchallenge, 0.188 ; openmic, 0.174 ; giveaway, 0.145 ; artstorm, 0.144 ; photobomb, 0.122 ; sports, 0.111 ; newbiegames, 0.109 ; monomad, 0.108 ; design, 0.104 ; thealliance, 0.101
33	technology, 3.760	science, 0.351 ; business, 0.155 ; steemstem, 0.120 ; steemdunk, 0.114
34	inspiration, 2.517	philosophy, 0.425 ; onequality, 0.360 ; quote, 0.345 ; spirituality, 0.155
35	animals, 7.821	animalphotography, 1.388 ; cats, 0.852 ; pets, 0.672 ; dogs, 0.486 ; cat, 0.397 ; dog, 0.370 ; dailypetphotography, 0.319 ; homesteading, 0.311 ; cute, 0.288 ; birds, 0.288 ; wildlife, 0.154 ; pet, 0.136 ; caturday, 0.134 ; monomad, 0.102
36	games, 3.548	start, 1.206 ; steemgar, 1.195 ; steemplayroom, 1.194 ; gameplay, 0.441 ; gamer, 0.438 ; letsplay, 0.407
37	story, 3.813	fiction, 0.127
38	photo, 4.016
39	cryptocurrency, 7.017	ethereum, 0.536 ; altcoin, 0.247 ; trading, 0.239 ; altcoins, 0.166 ; eos, 0.102
40	lifestyle, 5.566	fitness, 0.319 ; fashion, 0.253 ; photocontest, 0.200 ; onequality, 0.187 ; steemdunk, 0.131 ; dlive-video, 0.112 ; beach, 0.104
41	game, 5.256	games, 0.121
42	community, 3.597	promo-steem, 0.647 ; curation, 0.432 ; teammalaysia, 0.146 ; freedom, 0.133 ; challenge30days, 0.132
43	history, 2.932	culture, 0.144
44	money, 5.134	trading, 0.260 ; finance, 0.258 ; altcoins, 0.224 ; business, 0.183 ; investing, 0.153 ; cryptonews, 0.134 ; stocks, 0.118
45	landscapephotography, 6.021	photocircle, 0.470 ; landscape, 0.364 ; yourluckyphotos, 0.225 ; photocontest, 0.176 ; newbieresteemday, 0.127 ; naturephotography, 0.126 ; mountains, 0.119 ; photofriend, 0.110 ; photomatic, 0.107 ; sunset, 0.106
46	cn, 4.038	cn-reader, 0.877 ; lizhi, 0.171 ; teammalaysia, 0.155 ; cn-book, 0.155 ; cn-malaysia, 0.153 ; book, 0.136 ; stats, 0.123
47	news, 4.077	geme, 0.144 ; comedy, 0.132
48	blog, 7.167
49	steemit, 7.670
50	dlive-broadcast, 7.562	dlive, 7.505 ; Dlive, 0.504 ; DLiveStar, 0.502 ; DliveStar, 0.323 ; Dlivestar, 0.267 ; fortnite, 0.259 ; dliver, 0.244 ; live, 0.237 ; stream, 0.230 ; dlivegaming, 0.214 ; Dunite, 0.201 ; learning, 0.200 ; of, 0.196 ; DLive, 0.195 ; , 0.182 ; Gaming, 0.174 ; DUnite, 0.171 ; steemgc, 0.149 ; pubg, 0.149 ; Fortnite, 0.146 ; dlivecommunity, 0.132 ; dliverewards, 0.127 ; Pubg, 0.113 ; csgo, 0.109 ; dunite, 0.103 ; league, 0.103
51	steepshot, 6.456
52	steemph, 4.632	philippines, 0.840 ; cebu, 0.312 ; steemph-antipolo, 0.220 ; steemitachievers, 0.214 ; teardrops, 0.201 ; pilipinas, 0.168 ; tilphilippines, 0.165 ; dailyfoodphotography, 0.155 ; steemitdavao, 0.155
53	smartphonephotography, 6.191	yourluckyphotos, 0.130 ; smartphotography, 0.110
54	video, 4.522	youtube, 0.313 ; recipe, 0.201 ; people, 0.156 ; vlog, 0.155 ; dmania, 0.143
55	aceh, 7.711	acehmacro, 0.108
56	stach, 5.362	onequality, 0.921 ; promo-steem, 0.347 ; teardrops, 0.163 ; steem9ja, 0.122 ; sndbox, 0.114 ; joinsteemit, 0.100
57	flowers, 4.427	petals, 0.623 ; garden, 0.407 ; gardening, 0.239 ; plants, 0.162 ; summer, 0.130 ; spring, 0.124
58	philippines, 4.744	steempress, 1.228 ; cebu, 0.289 ; ulog, 0.166 ; redfish, 0.125
59	castellano, 4.063	venezuela, 1.064 ; literatura, 0.178 ; provenezuela, 0.114 ; proconocimiento, 0.105
60	wafrica, 5.362	reachout, 0.199
61	ico, 6.998	ethereum, 2.400 ; bounty, 0.349 ; crowdsale, 0.237 ; tokensale, 0.233 ; eth, 0.217 ; token, 0.153 ; many, 0.132 ; investment, 0.128 ; blokchain, 0.116 ; investing, 0.112 ; review, 0.105
62	tr, 3.658	cointurk, 1.568 ; destektr, 0.488 ; trliste, 0.477 ; kusadasi, 0.323 ; gif, 0.258 ; failarmy, 0.241 ; hede-io, 0.223 ; information, 0.175 ; anadolu, 0.108
63	article, 6.533	myanmar, 0.150
64	motivation, 4.328	onequality, 0.216 ; quote, 0.191 ; success, 0.189 ; quotes, 0.163
65	poetry, 4.547	poem, 0.868 ; steemitschoolpoetry, 0.209 ; poetsunited, 0.138 ; post, 0.117
66	kr, 7.209	coinkorea, 0.766 ; kr-life, 0.548 ; kr-writing, 0.483 ; kr-daily, 0.340 ; kr-event, 0.301 ; kr-art, 0.240 ; kr-pen, 0.236 ; kr-coin, 0.228 ; kr-diary, 0.157 ; tooza, 0.135 ; muksteem, 0.132 ; kr-news, 0.113 ; kr-travel, 0.113 ; kr-gazua, 0.112
67	writing, 6.765	fiction, 0.423 ; freewrite, 0.125 ; myanmar, 0.123
68	christian-trail, 3.038	steemchurch, 3.035 ; christianity, 2.615 ; religion, 1.764 ; bible, 0.559 ; spirituality, 0.385 ; communitynews, 0.304 ; faith, 0.255 ; flaminghelpers, 0.226 ; jesus, 0.192 ; gospel, 0.127 ; christian, 0.116 ; god, 0.104 ; ghana, 0.101
69	esteem, 10.849	sevendaybnwchallenge, 0.133
70	world, 5.584	creativity, 1.384 ; creative, 1.320 ; adventure, 0.738 ; usa, 0.292 ; amazing, 0.242 ; immigration, 0.194 ; of, 0.165 ; future, 0.155 ; archive, 0.142 ; people, 0.137 ; invention, 0.129 ; diy, 0.117 ; film, 0.113 ; cup, 0.101
71	good-karma, 4.972	time, 0.751 ; religious, 0.243 ; information, 0.240 ; steemdunk, 0.112 ; mso, 0.101
72	colorchallenge, 3.949	thursdaygreen, 0.127 ; wednesdayyellow, 0.102
73	animal, 5.792	animalphotography, 0.563 ; cat, 0.347 ; dog, 0.161 ; acehmacro, 0.111
74	free, 2.315	minnowsupport, 2.190 ; giveaway, 2.006 ; resteem, 0.145
75	hot, 4.229	trending, 3.951 ; promoted, 2.900 ; girls, 0.181 ; sexy, 0.176 ; asmr, 0.165 ; viral, 0.123
76	family, 4.035	happy, 0.279 ; muzakirpb, 0.188 ; freedom, 0.147 ; english, 0.135 ; culture, 0.130 ; familyprotection, 0.124 ; kids, 0.121 ; education, 0.120 ; ulog, 0.116 ; children, 0.104
77	new, 4.442
78	science, 3.635	education, 3.413 ; steemstem, 0.751 ; steemiteducation, 0.622 ; people, 0.607 ; space, 0.224 ; sndbox, 0.159 ; psychology, 0.159 ; physics, 0.137 ; philosophy, 0.119 ; stemng, 0.118
79	deutsch, 4.702	steemit-austria, 0.557 ; german, 0.161 ; politik, 0.113 ; austria, 0.100
80	movie, 3.429	film, 2.223 ; review, 1.223 ; streaming, 0.804 ; films, 0.677 ; regarder, 0.661 ; movies, 0.539 ; ita, 0.297 ; comedy, 0.215 ; vn, 0.173 ; download, 0.162 ; full, 0.140 ; trailer, 0.127 ; drama, 0.126 ; avengers, 0.126 ; marvel, 0.110 ; romance, 0.108 ; online, 0.101
81	natural, 5.424	photografy, 0.289 ; healthy, 0.131 ; fruit, 0.116
82	minnowsupport, 3.889	investment, 3.589 ; bid-bot, 2.870 ; bid-bots, 2.869
83	gaming, 7.295	steemgc, 0.375 ; fortnite, 0.299 ; gameplay, 0.289 ; steemcraft, 0.281 ; letsplay, 0.264 ; gamer, 0.189 ; review, 0.142 ; votegame, 0.118 ; gamersunited, 0.113 ; ps4, 0.101
84	jjangjjangman, 5.574	muksteem, 0.517 ; kr-life, 0.419 ; kr-event, 0.308 ; kr-gazua, 0.186 ; kr-funfun, 0.184 ; dev, 0.156 ; it-news, 0.154 ; kr-travel, 0.144 ; kr-diary, 0.136 ; sharehows, 0.126 ; kr-overseas, 0.122 ; kr-youth, 0.121 ; kr-series, 0.109 ; kr-art, 0.107 ; kr-food, 0.102 ; kr-easy, 0.101
85	airdrop, 3.320	bounty, 1.363 ; token, 0.684 ; free, 0.236 ; airdrops, 0.203 ; upvote, 0.135 ; myanmar, 0.117 ; coin, 0.108
86	spanish, 5.171	venezuela, 0.284 ; steemitvenezuela, 0.120 ; talentclub, 0.114
87	sports, 2.641	football, 1.473 ; sport, 0.607 ; basketball, 0.487 ; soccer, 0.371 ; daily, 0.327 ; nba, 0.219 ; betting, 0.206 ; steemsports, 0.196
88	indonesia, 8.426	ksi, 0.743 ; garudakita, 0.225 ; promo-steem, 0.190 ; bogor, 0.157
89	fun, 4.964	picture, 0.131 ; meme, 0.119 ; humor, 0.118
90	photofeed, 7.138	photocircle, 1.444 ; photomatic, 0.722 ; portraitphotography, 0.471 ; yourluckyphotos, 0.412 ; goldenhourphotography, 0.297 ; streetphotography, 0.291 ; monomad, 0.261 ; bescouted, 0.259 ; architecturalphotography, 0.232 ; vehiclephotography, 0.184 ; cityscapephotography, 0.168 ; animalphotography, 0.137 ; photocontest, 0.130 ; photocontests, 0.130 ; petals, 0.116 ; bwphotocontest, 0.110
91	nigeria, 4.588	onequality, 0.432 ; africa, 0.298 ; ulog, 0.166 ; genesisproject, 0.126
92	dlivestar, 5.661	steemgc, 1.786 ; dunite, 1.679 ; dliver, 1.260 ; dlivegaming, 1.103 ; dlive, 0.993 ; dlive-broadcast, 0.942 ; fr, 0.492 ; dlive24hour, 0.327 ; fortnite, 0.263 ; lovedlive, 0.239 ; dliverewards, 0.236 ; dlivestarbooster, 0.211 ; adsactly, 0.165 ; dlivebroadcast, 0.143 ; dota2, 0.142 ; dlivecommunity, 0.125 ; bgame, 0.121 ; game, 0.117 ; dgamer, 0.107 ; dlivestreaming, 0.107 ; livestream, 0.105 ; teamdlive, 0.102 ; upvote, 0.101 ; dlivebooster, 0.100
93	steemgigs, 3.356	teardrops, 3.212 ; ulog, 0.945 ; surpassinggoogle, 0.854 ; ulogs, 0.453 ; thai, 0.405 ; steem-untalented, 0.199 ; gratefulvibes, 0.178 ; untalented-steemgigs, 0.119 ; sndbox, 0.111 ; philippines, 0.108 ; thaipowerup, 0.101
94	funny, 6.766	comedy, 0.873 ; geme, 0.350 ; humor, 0.331 ; memes, 0.237 ; gif, 0.184 ; joke, 0.179 ; jokes, 0.159 ; punchline, 0.155 ; lol, 0.124
95	crypto, 7.408	trading, 0.800 ; altcoin, 0.755 ; ethereum, 0.482 ; btc, 0.425 ; currency, 0.214 ; investing, 0.178 ; eos, 0.144 ; market, 0.120 ; eth, 0.119
96	india, 3.745	pictures, 0.162 ; indiaunited, 0.152 ; venezuela, 0.144 ; design, 0.117 ; culture, 0.101
97	dsound, 3.321	dsound-original, 1.956 ; ftlob, 0.435 ; musicvoter, 0.416 ; steem-music, 0.314 ; dsound-cover, 0.312 ; dsound-music, 0.306 ; podcast, 0.303 ; music, 0.286 ; dsound-podcast, 0.267 ; dsound-electronic, 0.227 ; dsound-dsound, 0.216 ; dsound-pop, 0.208 ; dsound-instrumental, 0.145 ; dsound-hiphop, 0.138 ; instrumental, 0.105
98	nsfw, 2.889	sexy, 1.681 ; porn, 1.002 ; sex, 0.678 ; nude, 0.497 ; girl, 0.420 ; girls, 0.311 ; hot, 0.276 ; adult, 0.274 ; woman, 0.250 ; dporn, 0.205 ; hentai, 0.202 ; japanese, 0.182 ; hardcore, 0.172 ; model, 0.165 ; anime, 0.157 ; pussy, 0.155 ; boobs, 0.138 ; asmr, 0.137 ; erotica, 0.117 ; manga, 0.106 ; ass, 0.105
99	politics, 3.952	businessandcommerce, 0.971 ; hollywoodandmovies, 0.971 ; musicandentertainment, 0.971 ; scienceandtechnology, 0.971 ; freedom, 0.442 ; informationwar, 0.427 ; trump, 0.175 ; anarchy, 0.164 ; conspiracy, 0.160 ; liberty, 0.109