I'm trying to train my Wit.ai bot in order to recognize the first name of someone. I'm not very sure if I well understand how the NLP works so I'll give you an example.
I defined a lot of expressions like "My name is XXXX", "Everybody calls me XXXX"
In the "Understanding" table I added an entity named "contact_name" and I add almost 50 keywords like "Michel, John, Mary...".
I put the trait as "free-text" and "keywords".
I'm not sure if this process is correctly. So, I ask you:
does it matter the context like "My name is..." for the NLP? I mean...will it help the bot to predict that after this expression probably a fist name will come on?
is that right to add like 50 values to an entity or it's completly wrong?
what do you suggest as a training process in order to get the first name of someone?
You have done it right by keeping the entity's search strategy as "free-text" and "Keywords". But Adding keywords examples to the entity doesn't make any sense because a person's name is not a keyword.
So, I would recommend a training strategy which is as follows:
Create various templates of the message like, "My name is XYZ", "I am XYZ", "This is XYZ" etc. (all possible introduction messages you could think of)
Remove all keywords and expressions for the entity you created and add these two keywords:
"a b c d e f g h i j k l m n o p q r s t u v w x y z"
"XYZ" (can give any name but maintain this name same for validating the templates)
In the 'Understanding' tab enter the messages and extract the name into the entity ("contact_name" in your case) and validate them
Similarly, validate all message templates keeping the name as "XYZ"
After you have done this for all templates your bot will be able to recognise any name in a given template of the message
The logic behind this is your entity is a free-text and keyword which means it first tries to match the keyword if not matched it tries to find the word in the same position of the templates. Keeping the name same for validations helps to train the bot with the templates and learn the position where the name will be usually found.
Hope this works. I have tried this and worked for me. I am not sure how bot trains in background. I recommend you to start a new app and do this exercise.
Comment if there is any problem.
wit.ai has a pre-trained entity extraction method called wit/contact, which
Captures free text that's either the name or a clear reference to a
person, like "Paul", "Paul Smith", "my husband", "the dentist".
It works good even without any training data.
To read about the method refer to duckling.
Related
I am using cloud google vision API to extract text from Aadhaar and PAN. How can I get exact user details like name, father's name, and address?
Raw Data
ଭାରତ ସରକାର
Government of India
ଜିତ୍ୟାନନ୍ଦ ଖେମୁକୁ
NITYANANDA KHEMUDU
ପିତା : ସୀତାରାମ ଖେମୁକୁ
Father: Sitaram Khemudu
ଜନ୍ମ ତାରିଖ / DOB : 01.07.1999
ପୁରୁଷ / Male
ମୋ ଆଧାର, ମୋ ପରିଚୟ
I have built 5-6 OCR till date like aadhar, pan, ITR, Driving Linces etc., using google cloud vision API, I think you are looking for response like
{"pan_card_no":"ECXXXXXX123",
"name":"fshksj"
}
to get such response you need to built your own logic, here are some logic's i can share with you
Perform OCR on your document using Google_cloud_vision API and store that response into one array (Goggle gives logic line by line)
Like in above case if you want to grab DOB first you can build logic like i) if "DOB" in (list of item) then grab the numeric values
To get the name what you can do is dropping the unnecessary items from list by if using if condition like (if "India" in i) or (if i.isdigit()) then drop it likewise you can drop the unnesseary items from main list to get the Name
to grab the Address what you can do is, 95% of the time address come with pincode at last, so what you can do is treat pincode as a last index of address and look of "Address" kind of keyword then add all the elements from "Add keyword index" to "pincode index" ( this can be easily done in list) to validate whether the pincode is valid or not you can use library like Pyzipin
There are multiple conditions that you can use, above are the very basic one i mentioned, if you need any specific logic then then you can ask me
Imagine I have a report, a letter actually, which I need to translate to several languages. I have created a greeting field in the form which is filled programatically by an onchange event method.
if self.partner_id.gender == 'female':
self.letter_greeting = _('Dear %s %s,') % ( # the translation should be "Estimada"
self.repr_recipient_id.title.shorcut, surname
)
elif self.partner_id.gender == 'male':
self.letter_greeting = _('Dear %s %s,') % ( # translation "Estimado"
self.repr_recipient_id.title.shorcut, surname
)
else:
self.letter_greeting = _('Dear %s %s,') % ( # translation: "Estimado/a"
self.partner_id.title.shorcut, surname
)
In that case the word Dear should be translated to different Spanish translations depending on which option is used, this is because we use different termination depending on the gender. Exporting the po file I found that all the options are altogether, that make sense because almost all the cases the translations will be the same, but not in this case:
#. module: custom_module
#: code:addons/custom_module/models/sale_order.py:334
#: code:addons/custom_module/models/sale_order.py:338
#: code:addons/custom_module/models/sale_order.py:342
#, python-format
msgid "Dear %s %s,"
msgstr "Dear %s %s,"
Solutions I can apply directly
Put all the terms in different entries to avoid the same translation manually every time I need to update the po file. This can be cumbersome if you have many different words with that problem. If I do it and I open the file with poedit, this error appears: duplicate message definition
Put all the possible combinations with slashes, this is done y some other parts of Odoo. For both gender would be:
#. module: stock
#: model:res.company,msg:stock.res_company
msgid "Dear"
msgstr "Estimado/a"
This is just an example. I can think of many words that look the same in English, but they use different spelling or meanings in other languages depending on the context.
Possible best solutions
I don't know if Odoo know anything aboutu the context of a word to know if it was already translated or not. Adding a context manually could solve the problem, at least for words with different meanings.
The nicest solution would be to have a parameter to the translation module to make sure that the word is exported as an isolated entry for that especific translation.
Do you think that I am giving to it too much importance haha? Do you know if there is any better solution? Why is poedit not taking into account that problem at all?
I propose an extension of models res.partner.title and res.partner.
res.partner.title should get a translateable field for saving salutation prefixes like 'Dear' or 'Sehr geehrter' (German). Maybe it's worth to get something about genders, too, but i won't get into detail here about that.
You probably want to show the configuring user an example like "Dear Mr. Name" or something like that. A computed field should work.
On res.partner you should just implement either a computed field or just a method to get a full salutation for a partner record.
To some degree this is a linguistics problem. I believe the best solution would be to use a different "Source Language", one made up of keys, and then have English as another Translation. The word "Dear" in English does not have a gender context (and typically, much of English doesn't), while the word "Estimado" in Spanish does. The translation from that Spanish word to English is more appropriately "Masculine Dear." Therefore, using keys as your source language, you would have this:
SourceText (EnglishDescription) -> Translation (English) -> Translation (Spanish)
DearMasculine -> Dear -> Estimado
DearFeminine -> Dear -> Estimada
DearNuetral -> Dear -> Estimado/a
I am using premium account (not sandbox) for data collection.
I want to collect:
All tweets in English that contain ‘china’ or ‘chinese’ that are user geolocated to US and not geolocated at tweet level, excluding all retweets
All tweets in English that contain ‘china’ or ‘chinese’ that are user geolocated to ‘Minnesota’ and not geolocated at tweet level, excluding all retweets
The code is as follows:
premium_search_args = load_credentials('twitter_API.yaml',
yaml_key ='search_tweets_premium_api', env_overwrite=False)
# keywords for the search
# key word 1
keywords = '(China OR Chinese) lang:en profile_country:US -place_country:US -is:retweet'
# key word 2
keywords = '(China OR Chinese) lang:en -place_country:US profile_region:"Minnesota" -is:retweet'
# define search rule
rule = gen_rule_payload(keywords,from_date='2019-12-01',
to_date='2019-12-10',results_per_call=500)
# create result stream and print before start
rs = ResultStream(rule_payload=rule, max_results=1250000,
**premium_search_args)
My problems are that:
For the first one, a large portion of the results I get didn’t satisfy the query. First, some don’t have Profile Geo enrichment, i.e. user.derived.locations attribute is not in the user object. Second, if it is, a lot don’t have country code US, i.e. they are identified to other countries.
For the second one, the result I get from this method is a smaller subset of the results I can get from 1). That is, when I filter all tweets user geolocated to Minnesota (by user.derived.locations.region) from profile_country:US, it gives a larger sample than using profile_region:“Minnesota”. A considerable amount of data is missing using this method.
I have tried several times but it seems that user geolocation operators don’t work exactly what I want. Does anyone has any idea why this is the case? I would very much appreciate any answers/suggestions/comments.
Thank you!
I have a relation called conversations_grouped made up of bags of tuples of varying sizes, like so:
DUMP conversations_grouped:
...
({(L194),(L195),(L196),(L197)})
({(L198),(L199)})
({(L200),(L201),(L202),(L203)})
({(L204),(L205),(L206)})
({(L207),(L208)})
({(L271),(L272),(L273),(L274),(L275)})
({(L276),(L277)})
({(L280),(L281)})
({(L363),(L364)})
({(L365),(L366)})
({(L666256),(L666257)})
({(L666369),(L666370),(L666371),(L666372)})
({(L666520),(L666521),(L666522)})
Each L[0-9]+ is a tag corresponding to a string. For example, L194 might be "Hello, how are you doing?" and L195 might be "fine, how are you?". This correspondence is maintained by a map called line_map. Here's a sample:
DUMP line_map;
...
([L666324#Do you think she might be interested in someone?])
([L666264#Well that's typical of Her Majesty's army. Appoint an engineer to do a soldier's work.])
([L666263#Um. There are rumours that my Lord Chelmsford intends to make Durnford Second in Command.])
([L666262#Lighting COGHILL' 5 cigar: Our good Colonel Dumford scored quite a coup with the Sikali Horse.])
([L666522#So far only their scouts. But we have had reports of a small Impi farther north, over there. ])
([L666521#And I assure you, you do not In fact I'd be obliged for your best advice. What have your scouts seen?])
([L666520#Well I assure you, Sir, I have no desire to create difficulties. 45])
([L666372#I think Chelmsford wants a good man on the border Why he fears a flanking attack and requires a steady Commander in reserve.])
([L666371#Lord Chelmsford seems to want me to stay back with my Basutos.])
([L666370#I'm to take the Sikali with the main column to the river])
([L666369#Your orders, Mr Vereker?])
([L666257#Good ones, yes, Mr Vereker. Gentlemen who can ride and shoot])
([L666256#Colonel Durnford... William Vereker. I hear you 've been seeking Officers?])
What I'm trying to do now is parse through each line and replace the L[0-9]+ tags with their corresponding text from line_map. Is it possible to make references to line_map from within a Pig FOREACH statement, or is there something else I have to do?
The first issue with this is that in a map the key must be a quoted string. So you can't use a schema value to access the map. E.G. This will not work.
C: {foo: chararray, M: [value:chararray]}
D = FOREACH C GENERATE M#foo ;
The solution that comes to mind is to FLATTEN conversations_grouped. Then do a join between conversations_grouped and line_map on the L[0-9]+ tag. You'll probably want to project out some of the extra fields (like the L[0-9]+ tag after the join) to make the next step faster. After that you'll have to regroup the data, and massage it into the correct format.
This won't work unless each bag has it's own unique ID for the regrouping, but if each of the L[0-9]+ tags appear in only one bag (conversation) you can use this to create a unique id.
-- A is dumped conversations_grouped
B = FOREACH A {
-- Pulls out an element from the bag to use as the id
id = LIMIT tags 1 ;
-- Flattens B into id, tag form. Each group of tags will have the same id.
GENERATE FLATTEN(id), FLATTEN(tags) ;
}
The schema and output for B is:
B: {id: chararray,tags::tag: chararray}
(L194,L194)
(L194,L195)
(L194,L196)
(L194,L197)
(L198,L198)
(L198,L199)
(L200,L200)
(L200,L201)
(L200,L202)
(L200,L203)
(L204,L204)
(L204,L205)
(L204,L206)
(L207,L207)
(L207,L208)
(L271,L271)
(L271,L272)
(L271,L273)
(L271,L274)
(L271,L275)
(L276,L276)
(L276,L277)
(L280,L280)
(L280,L281)
(L363,L363)
(L363,L364)
(L365,L365)
(L365,L366)
(L666256,L666256)
(L666256,L666257)
(L666369,L666369)
(L666369,L666370)
(L666369,L666371)
(L666369,L666372)
(L666520,L666520)
(L666520,L666521)
(L666520,L666522)
Assuming that the tags are unique, the rest is done like:
-- A2 is line_map, loaded in tag/message pairs instead of a map
-- Joins conversations_grouped and line_map on tag
C = FOREACH (JOIN B by tags::tag, A2 by tag)
-- This generate removes the tag
GENERATE id, message ;
-- Regroups C on the id created in B
D = FOREACH (GROUP C BY id)
-- This step limits the output to just messages
GENERATE C.(message) AS messages ;
Schema and output from D:
D: {messages: {(A2::message: chararray)}}
({(Colonel Durnford... William Vereker. I hear you 've been seeking Officers?),(Good ones, yes, Mr Vereker. Gentlemen who can ride and shoot)})
({(Your orders, Mr Vereker?),(I'm to take the Sikali with the main column to the river),(Lord Chelmsford seems to want me to stay back with my Basutos.),(I think Chelmsford wants a good man on the border Why he fears a flanking attack and requires a steady Commander in reserve.)})
({(Well I assure you, Sir, I have no desire to create difficulties. 45),(And I assure you, you do not In fact I'd be obliged for your best advice. What have your scouts seen?),(So far only their scouts. But we have had reports of a small Impi farther north, over there. )})
NOTE: If at worst, (the L[0-9]+ tags aren't unique) you can give each line of the input file(s) a sequential, integer id before you load it into pig.
UPDATE: If you are using pig 0.11, then you can also use the RANK operator.
I need to write a text parser for the education domain which can extract out the information like institute, location, course etc from the free text.
Currently i am doing it through lucene, steps are as follows:
Index all the data related to institute, courses and location.
Making shingles of the free text and searching each shingle in location, course and institute index dir and then trying to find out which part of text represents location, course etc.
In this approach I am missing lot of cases like B.tech can be written as btech, b-tech or b.tech.
I want to know is there any thing available which can do all these kind of things, I have heard about Ling-pipe and Gate but don't know how efficient they are.
You definitely need GATE. GATE has 2 main most frequently used features (among thousands others): rules and dictionaries. Dictionaries (gazetteers in GATE's terms) allow you to put all possible cases like "B.tech", "btech" and so on in a single text file and let GATE find and mark them all. Rules (more precisely, JAPE-rules) allow you to define patterns in text. For example, here's pattern to catch MIT's postal address ("77 Massachusetts Ave., Building XX, Cambridge MA 02139"):
{Token.kind == number}(SP){Token.orth == uppercase}(SP){Lookup.majorType == avenue}(COMMA)(SP)
{Token.string == "Building"}(SP){Token.kind == number}(COMMA)(SP)
{Lookup.majorType == city}(SP){Lookup.majorType == USState}(SP){Token.kind == number}
where (SP) and (COMMA) - macros (just to make text shorter), {Somthing} - is annotation, , {Token.kind == number} - annotation "Token" with feature "kind" equal to "number" (i.e. just number in the text), {Lookup} - annotation that captures values from dictionary (BTW, GATE already has dictionaries for such things as US cities). This is quite simple example, but you should see how easily you can cover even very complicated cases.
I didn't use Lucene but in your case I would leave different forms of the same keyword as they are and just hold a link table or such. In this table I'd keep the relation of these different forms.
You may need to write a regular expression to cover each possible form of your vocabulary.
Be careful about your choice of analyzer / tokenizer, because words like B.tech can be easily split into 2 different words (i.e. B and tech).
You may want to check UIMA. As Lingpipe and Gate, this framework features text annotation, which is what you are trying to do. Here is a tutorial which will help you write an annotator for UIMA:
http://uima.apache.org/d/uimaj-2.3.1/tutorials_and_users_guides.html#ugr.tug.aae.developing_annotator_code
UIMA has addons, in particular one for Lucene integration.
You can try http://code.google.com/p/graph-expression/
example of Adress parsing rules
GraphRegExp.Matcher Token = match("Token");
GraphRegExp.Matcher Country = GraphUtils.regexp("^USA$", Token);
GraphRegExp.Matcher Number = GraphUtils.regexp("^\\d+$", Token);
GraphRegExp.Matcher StateLike = GraphUtils.regexp("^([A-Z]{2})$", Token);
GraphRegExp.Matcher Postoffice = seq(match("BoxPrefix"), Number);
GraphRegExp.Matcher Postcode =
mark("Postcode", seq(GraphUtils.regexp("^\\d{5}$", Token), opt(GraphUtils.regexp("^\\d{4}$", Token))))
;
//mark(String, Matcher) -- means creating chunk over sub matcher
GraphRegExp.Matcher streetAddress = mark("StreetAddress", seq(Number, times(Token, 2, 5).reluctant()));
//without new lines
streetAddress = regexpNot("\n", streetAddress);
GraphRegExp.Matcher City = mark("City", GraphUtils.regexp("^[A-Z]\\w+$", Token));
Chunker chunker = Chunkers.pipeline(
Chunkers.regexp("Token", "\\w+"),
Chunkers.regexp("BoxPrefix", "\\b(POB|PO BOX)\\b"),
new GraphExpChunker("Address",
seq(
opt(streetAddress),
opt(Postoffice),
City,
StateLike,
Postcode,
Country
)
).setDebugString(true)
);
B.tech can be written as btech, b-tech or b.tech
Lucene will let you do fuzzy searches based on the Levenshtein Distance. A query for roam~ (note the ~) will find terms like foam and roams.
That might allow you to match the different cases.