{"description": "", "language": {"code": "en", "dir": "ltr", "name": "English"}, "metadata": {}, "note": "", "resource_uri": "/api2/partners/videos/ZElJjOYQduxT/languages/en/subtitles/", "site_url": "http://www.amara.org/videos/ZElJjOYQduxT/en/100268/", "sub_format": "json", "subtitles": [{"end": 4920, "meta": {"new_paragraph": true}, "position": 1, "start": 0, "text": "In our baseline algorithm, we used all the
words to do sentiment analysis. But"}, {"end": 10225, "meta": {"new_paragraph": false}, "position": 2, "start": 4920, "text": "there's lots of information about which
words have which kind of sentiment that we"}, {"end": 15210, "meta": {"new_paragraph": false}, "position": 3, "start": 10225, "text": "might want to make use of in addition.
These are called sentiment lexicons. One"}, {"end": 20386, "meta": {"new_paragraph": false}, "position": 4, "start": 15210, "text": "sentiment lexicon is the General Inquirer.
It's been around for quite a long time."}, {"end": 25755, "meta": {"new_paragraph": false}, "position": 5, "start": 20386, "text": "Since 1966. And I pointed you here at the
homepage and the list of categories for"}, {"end": 30740, "meta": {"new_paragraph": false}, "position": 6, "start": 25755, "text": "the General Inquirer and you can see the
spreadsheet that lists all the words."}, {"end": 39671, "meta": {"new_paragraph": false}, "position": 7, "start": 31160, "text": "Let's go look at the General Inquirer. So,
here's the general inquiry categories. And"}, {"end": 50298, "meta": {"new_paragraph": false}, "position": 8, "start": 39671, "text": "here for example, you can see This a
little larger. Here you can see positive"}, {"end": 55906, "meta": {"new_paragraph": false}, "position": 9, "start": 50298, "text": "and negative lists of words. Let's go look
at one of them. Let's look at the positive"}, {"end": 64273, "meta": {"new_paragraph": false}, "position": 10, "start": 55906, "text": "list. So here's some list of words,
ability, abide, able, abound, absolve, and"}, {"end": 74705, "meta": {"new_paragraph": false}, "position": 11, "start": 64273, "text": "so on. So here's positive words from this
intimate list. So, in addition to positive"}, {"end": 79690, "meta": {"new_paragraph": false}, "position": 12, "start": 74705, "text": "and negative the general [inaudible] also
has lots of other lists of kinds of"}, {"end": 84675, "meta": {"new_paragraph": false}, "position": 13, "start": 79690, "text": "classes or words, strong versus weak,
overstated versus understated, words about"}, {"end": 89916, "meta": {"new_paragraph": false}, "position": 14, "start": 84675, "text": "pleasure, cognitive orientation, all sorts
of other words. So we can use it not just"}, {"end": 94900, "meta": {"new_paragraph": false}, "position": 15, "start": 89916, "text": "for the simple polarity questions that
we're talking about right now, but more"}, {"end": 100315, "meta": {"new_paragraph": false}, "position": 16, "start": 94900, "text": "rich sentiment kind of questions. Luke,
the linguistic inquiry and word count, is"}, {"end": 105910, "meta": {"new_paragraph": false}, "position": 17, "start": 100315, "text": "a, database developed by, James Pennybaker
and his colleagues. And here we have lots"}, {"end": 110932, "meta": {"new_paragraph": false}, "position": 18, "start": 105910, "text": "of classes, 70 classes of words talking
about negative emotion, positive emotion."}, {"end": 116145, "meta": {"new_paragraph": false}, "position": 19, "start": 110932, "text": "Words like bad or weird or hate. Cognitive
processes. All the pronouns. First person,"}, {"end": 120786, "meta": {"new_paragraph": false}, "position": 20, "start": 116145, "text": "second person pronouns. Words having to do
with negation, like no or never."}, {"end": 125681, "meta": {"new_paragraph": false}, "position": 21, "start": 120786, "text": "Quantifiers and so on. And Luke has a
small charge, but has a large list of"}, {"end": 132453, "meta": {"new_paragraph": false}, "position": 22, "start": 125681, "text": "words that people often use for sentient
analysis. [sound] The MPQA subjective, the"}, {"end": 140634, "meta": {"new_paragraph": false}, "position": 23, "start": 132453, "text": "Qs lexicon, by Wilson and Weebee and their
colleagues. Gives you again a list of, of"}, {"end": 145945, "meta": {"new_paragraph": false}, "position": 24, "start": 140634, "text": "polarity items, your positive words and
negative words. And words in this lexicon"}, {"end": 152305, "meta": {"new_paragraph": false}, "position": 25, "start": 145945, "text": "are also annotated for intensity, for how
strong or weak they are. Another polarity"}, {"end": 158238, "meta": {"new_paragraph": false}, "position": 26, "start": 152305, "text": "lexicon that's available on the wedding
web is, is Bing Leo's opinion lexicon and"}, {"end": 165120, "meta": {"new_paragraph": false}, "position": 27, "start": 158238, "text": "I pointed you at that. Again, another list
of positive or negative words. Finally, we"}, {"end": 170028, "meta": {"new_paragraph": false}, "position": 28, "start": 165120, "text": "haven't talked about Wordnet, we'll talk
about Wordnet later. But Wordnet is an"}, {"end": 174873, "meta": {"new_paragraph": false}, "position": 29, "start": 170028, "text": "online thesaurus and the entries in
Wordnet have also been labeled for their"}, {"end": 181736, "meta": {"new_paragraph": false}, "position": 30, "start": 174873, "text": "polarity. So for example, the, word
estimable, may be computed, or estimated,"}, {"end": 188411, "meta": {"new_paragraph": false}, "position": 31, "start": 182010, "text": "is labeled as, neither positive or
negative, but objective, In the"}, {"end": 193426, "meta": {"new_paragraph": false}, "position": 32, "start": 188411, "text": "Sentiwordnet, but the word, estimable,
meaning deserving of respect or high"}, {"end": 198833, "meta": {"new_paragraph": false}, "position": 33, "start": 193426, "text": "regard, that's a mostly positive, although
sometimes not, sometimes objective word."}, {"end": 204109, "meta": {"new_paragraph": false}, "position": 34, "start": 198833, "text": "So we can see that each word gets a label
for how positive, negative, or objective"}, {"end": 209711, "meta": {"new_paragraph": false}, "position": 35, "start": 204109, "text": "it is. And again you can go look at that
here, and you have to fill out a form to"}, {"end": 215005, "meta": {"new_paragraph": false}, "position": 36, "start": 209711, "text": "download that. And we'll talk about
Wordnet itself a bit later. [sound]. So"}, {"end": 220568, "meta": {"new_paragraph": false}, "position": 37, "start": 215005, "text": "each of these lexicons has different
things in it but they overlap in all"}, {"end": 226431, "meta": {"new_paragraph": false}, "position": 38, "start": 220568, "text": "talking about polarity. And Chris Potts
ran a lovely correlation between these,"}, {"end": 232671, "meta": {"new_paragraph": false}, "position": 39, "start": 226431, "text": "looking at whether the polarity of a word
is in multiple lexicons does it have the"}, {"end": 238459, "meta": {"new_paragraph": false}, "position": 40, "start": 232671, "text": "same polarity? And you can see that
sentiwordnet has a whole lot of different"}, {"end": 243603, "meta": {"new_paragraph": false}, "position": 41, "start": 238459, "text": "things going on in it, in general. If a
word is in multiple lexicons, the, the,"}, {"end": 249183, "meta": {"new_paragraph": false}, "position": 42, "start": 243793, "text": "the disagreements between the lexicons are
very small. So most of the time, for the,"}, {"end": 254446, "meta": {"new_paragraph": false}, "position": 43, "start": 249373, "text": "opinion lexicon, the general [inaudible]
lexicon. For [inaudible], for the MPQA. In"}, {"end": 259772, "meta": {"new_paragraph": false}, "position": 44, "start": 254446, "text": "general, for all of these, you find pretty
similar sentiments if you're just doing"}, {"end": 264845, "meta": {"new_paragraph": false}, "position": 45, "start": 259772, "text": "polarity. Obviously, if you're doing more
complicated things, then they're, they're"}, {"end": 273260, "meta": {"new_paragraph": false}, "position": 46, "start": 264845, "text": "gonna start to differ. You might want to
look at each of words and, and talk about,"}, {"end": 277934, "meta": {"new_paragraph": false}, "position": 47, "start": 273460, "text": "What the polarity is of words, like words
that we get from these lexicons, if we're"}, {"end": 282191, "meta": {"new_paragraph": false}, "position": 48, "start": 277934, "text": "looking at a database like IMDB. And
there's some nice research by Chris Potts"}, {"end": 286720, "meta": {"new_paragraph": false}, "position": 49, "start": 282191, "text": "looking at this, so he asked the question,
how likely is each word to appear in each"}, {"end": 290867, "meta": {"new_paragraph": false}, "position": 50, "start": 286720, "text": "sentiment class? And let's look at IMDB
where there's ten stars so, eh, a word"}, {"end": 295065, "meta": {"new_paragraph": false}, "position": 51, "start": 290867, "text": "could like, bad, could appear in one star
reviews with a certain count, two star"}, {"end": 299377, "meta": {"new_paragraph": false}, "position": 52, "start": 295069, "text": "reviews with a certain count, three star
and so on. Can we use that to understand"}, {"end": 304229, "meta": {"new_paragraph": false}, "position": 53, "start": 299556, "text": "what's the sentiment of that word bad just
by counting how often it occurs in"}, {"end": 309320, "meta": {"new_paragraph": false}, "position": 54, "start": 304229, "text": "different reviews of different stars and
as he pointed out, you can't just use raw"}, {"end": 314142, "meta": {"new_paragraph": false}, "position": 55, "start": 309320, "text": "counts. Here, this shows you along the X
axis you look at the different categories,"}, {"end": 318659, "meta": {"new_paragraph": false}, "position": 56, "start": 314142, "text": "so one star, two star, three star, ten
star. And here's the count of times that"}, {"end": 323232, "meta": {"new_paragraph": false}, "position": 57, "start": 318659, "text": "the word bad occurs, the adjective bad
occurs. And you can see it's mostly in one"}, {"end": 328035, "meta": {"new_paragraph": false}, "position": 58, "start": 323232, "text": "star reviews. But it actually occurs more
often in ten star reviews than in two star"}, {"end": 332437, "meta": {"new_paragraph": false}, "position": 59, "start": 328035, "text": "reviews. Which seems like a, a problem,
because bad should be a really negative"}, {"end": 337300, "meta": {"new_paragraph": false}, "position": 60, "start": 332437, "text": "word. But that's because. There are just a
lot more nine and ten star reviews or"}, {"end": 342186, "meta": {"new_paragraph": false}, "position": 61, "start": 337300, "text": "eight and ten star reviews than there are
two star reviews. It turns out that"}, {"end": 347474, "meta": {"new_paragraph": false}, "position": 62, "start": 342376, "text": "reviews tend to skew positive. So we can't
use the raw counts. Instead we're going to"}, {"end": 352148, "meta": {"new_paragraph": false}, "position": 63, "start": 347474, "text": "use the likelihood. We're gonna take the
frequency of the word in the class and"}, {"end": 356799, "meta": {"new_paragraph": false}, "position": 64, "start": 352148, "text": "then we're gonna divide it by the total
number of words in that class. And that"}, {"end": 361101, "meta": {"new_paragraph": false}, "position": 65, "start": 356799, "text": "will solve the problem of classes like
nine and ten in general being bigger."}, {"end": 365972, "meta": {"new_paragraph": false}, "position": 66, "start": 361101, "text": "Cause we're computing the likelihood
condition on the class size. Now an"}, {"end": 370000, "meta": {"new_paragraph": false}, "position": 67, "start": 365972, "text": "additional thing we often wanna do is
compare these numbers for different words."}, {"end": 374029, "meta": {"new_paragraph": false}, "position": 68, "start": 370000, "text": "Some words are very frequent. Some words
are less frequent. And, so to do these we"}, {"end": 378158, "meta": {"new_paragraph": false}, "position": 69, "start": 374029, "text": "wanna do a scale of likelihood where we
take the likelihood of the word given the"}, {"end": 381834, "meta": {"new_paragraph": false}, "position": 70, "start": 378158, "text": "class and just divide by the probability
of the word, so now if a word is"}, {"end": 386114, "meta": {"new_paragraph": false}, "position": 71, "start": 381834, "text": "particularly frequent or particularly rare
we can still compare them with each other."}, {"end": 390518, "meta": {"new_paragraph": false}, "position": 72, "start": 386114, "text": "So we can compute the scale of
likelihoods. And if we look at that, we"}, {"end": 396051, "meta": {"new_paragraph": false}, "position": 73, "start": 390518, "text": "can see here's the word good or the word
amazing, we can see that as for amazing,"}, {"end": 401654, "meta": {"new_paragraph": false}, "position": 74, "start": 396051, "text": "as review numbers, as the stars go up from
one star to ten stars, the word is much"}, {"end": 407052, "meta": {"new_paragraph": false}, "position": 75, "start": 401654, "text": "more frequently used as we get higher up.
For great similarly, though not as strong"}, {"end": 412488, "meta": {"new_paragraph": false}, "position": 76, "start": 407052, "text": "as amazing. For awesome, again we see a
very sharp, rise in its, in its use, in"}, {"end": 417550, "meta": {"new_paragraph": false}, "position": 77, "start": 412488, "text": "[inaudible] reviews. For good it turns out
to be used mostly in the middle. A good"}, {"end": 422675, "meta": {"new_paragraph": false}, "position": 78, "start": 417550, "text": "movie is not great, and it's not terrible,
and that's what we might expect from the"}, {"end": 428297, "meta": {"new_paragraph": false}, "position": 79, "start": 422675, "text": "word good. And similarly for, bad reviews.
So, again, a word like depressed, more"}, {"end": 433561, "meta": {"new_paragraph": false}, "position": 80, "start": 428297, "text": "often can occur in these negative reviews,
the one, two, and three reviews. A word"}, {"end": 438692, "meta": {"new_paragraph": false}, "position": 81, "start": 433561, "text": "like bad, again, occurring in these one,
two, and three star reviews. Terrible in"}, {"end": 443957, "meta": {"new_paragraph": false}, "position": 82, "start": 438692, "text": "these one, two, and three star reviews. So
we can look at the review-, along the X"}, {"end": 449154, "meta": {"new_paragraph": false}, "position": 83, "start": 443957, "text": "axis, the, the review quality, one through
ten. And along the Y axis, the scale"}, {"end": 455105, "meta": {"new_paragraph": false}, "position": 84, "start": 449154, "text": "likelihood to see the polarity of the
word. So Potter also asked, whether,"}, {"end": 461546, "meta": {"new_paragraph": false}, "position": 85, "start": 455356, "text": "whether we can see sentiment in other
words, then, good or bad. He looked at"}, {"end": 468154, "meta": {"new_paragraph": false}, "position": 86, "start": 461797, "text": "logical negation, words like no, not,
never and the NT as isn't and didn't and"}, {"end": 474615, "meta": {"new_paragraph": false}, "position": 87, "start": 468154, "text": "so on. Are those associated with negative
sentiment? Now, what he did again, is he"}, {"end": 480215, "meta": {"new_paragraph": false}, "position": 88, "start": 474615, "text": "counted negation of online reviews and
looked at them, how they, how they match"}, {"end": 488958, "meta": {"new_paragraph": false}, "position": 89, "start": 480215, "text": "the review rating. And, just like the word
bad. And, or the word terrible. You can"}, {"end": 494364, "meta": {"new_paragraph": false}, "position": 90, "start": 488958, "text": "see that the word, these negation words,
like no, not, never, occur more frequently"}, {"end": 499770, "meta": {"new_paragraph": false}, "position": 91, "start": 494364, "text": "in these low star reviews. Either in the
IMBD database, or other databases which"}, {"end": 505107, "meta": {"new_paragraph": false}, "position": 92, "start": 499770, "text": "are only from one to five stars. So in
these lexicons, we can see lots of words"}, {"end": 510445, "meta": {"new_paragraph": false}, "position": 93, "start": 505107, "text": "are positive and negative. But there might
be more sophisticated features, like,"}, {"end": 515235, "meta": {"new_paragraph": false}, "position": 94, "start": 510445, "text": "let's say this, this kind of logical
negation also is a good cue for a"}], "title": "Sentiment Lexicons", "version_no": 3, "version_number": 3, "video": "Sentiment Lexicons", "video_description": "", "video_title": "Sentiment Lexicons"}