Commit 435d1076 authored by J. Fernando Sánchez's avatar J. Fernando Sánchez
Browse files

Add headers and minor fixes

parent c4321dc5
......@@ -4,8 +4,16 @@ All notable changes to this project will be documented in this file.
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
## [Unreleased]
## [1.0.1]
### Added
* License headers
* Description for PyPI (setup.py)
### Changed
* The evaluation tab shows datasets inline, and a tooltip shows the number of instances
* The docs should be clearer now
## [1.0.0]
### Fixed
* Restored hash changing function in `main.js`
......
This diff is collapsed.
This diff is collapsed.
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Senpy in 1 minute\n",
"\n",
"This mini-tutorial only shows how to annotate with a service.\n",
"We will use the [demo server](http://senpy.gsi.upm.es), which runs some open source plugins for sentiment and emotion analysis."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Annotating with senpy is as simple as issuing an HTTP request to the API using your favourite tool.\n",
"This is just an example using curl:"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{\r\n",
" \"@context\": \"http://senpy.gsi.upm.es/api/contexts/YXBpL3NlbnRpbWVudDE0MD8j\",\r\n",
" \"@type\": \"Results\",\r\n",
" \"entries\": [\r\n",
" {\r\n",
" \"@id\": \"prefix:\",\r\n",
" \"@type\": \"Entry\",\r\n",
" \"marl:hasOpinion\": [\r\n",
" {\r\n",
" \"@type\": \"Sentiment\",\r\n",
" \"marl:hasPolarity\": \"marl:Positive\",\r\n",
" \"prov:wasGeneratedBy\": \"prefix:Analysis_1554389334.6431913\"\r\n",
" }\r\n",
" ],\r\n",
" \"nif:isString\": \"Senpy is awesome\",\r\n",
" \"onyx:hasEmotionSet\": []\r\n",
" }\r\n",
" ]\r\n",
"}"
]
}
],
"source": [
"!curl \"http://senpy.gsi.upm.es/api/sentiment140\" --data-urlencode \"input=Senpy is awesome\""
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"**Congratulations**, you've used your first senpy service!"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Here is the equivalent using the `requests` library:"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {
"scrolled": true
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{\n",
" \"@context\": \"http://senpy.gsi.upm.es/api/contexts/YXBpL3NlbnRpbWVudDE0MD9pbnB1dD1TZW5weStpcythd2Vzb21lIw%3D%3D\",\n",
" \"@type\": \"Results\",\n",
" \"entries\": [\n",
" {\n",
" \"@id\": \"prefix:\",\n",
" \"@type\": \"Entry\",\n",
" \"marl:hasOpinion\": [\n",
" {\n",
" \"@type\": \"Sentiment\",\n",
" \"marl:hasPolarity\": \"marl:Positive\",\n",
" \"prov:wasGeneratedBy\": \"prefix:Analysis_1554389335.9803226\"\n",
" }\n",
" ],\n",
" \"nif:isString\": \"Senpy is awesome\",\n",
" \"onyx:hasEmotionSet\": []\n",
" }\n",
" ]\n",
"}\n"
]
}
],
"source": [
"import requests\n",
"res = requests.get('http://senpy.gsi.upm.es/api/sentiment140',\n",
" params={\"input\": \"Senpy is awesome\",})\n",
"print(res.text)"
]
}
],
"metadata": {
"anaconda-cloud": {},
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.3"
},
"toc": {
"colors": {
"hover_highlight": "#DAA520",
"running_highlight": "#FF0000",
"selected_highlight": "#FFD700"
},
"moveMenuLeft": true,
"nav_menu": {
"height": "68px",
"width": "252px"
},
"navigate_menu": true,
"number_sections": true,
"sideBar": true,
"threshold": 4,
"toc_cell": false,
"toc_section_display": "block",
"toc_window_display": false
}
},
"nbformat": 4,
"nbformat_minor": 1
}
This diff is collapsed.
Advanced usage
--------------
.. toctree::
:maxdepth: 1
server-cli
conversion
commandline
development
Command line
============
Although the main use of senpy is to publish services, the tool can also be used locally to analyze text in the command line.
This is a short video demonstration:
.. image:: https://asciinema.org/a/9uwef1ghkjk062cw2t4mhzpyk.png
:width: 100%
:target: https://asciinema.org/a/9uwef1ghkjk062cw2t4mhzpyk
:alt: CLI demo
......@@ -130,6 +130,7 @@ html_theme_options = {
'github_user': 'gsi-upm',
'github_repo': 'senpy',
'github_banner': True,
'sidebar_collapse': True,
}
......
Conversion
----------
Automatic Model Conversion
--------------------------
Senpy includes experimental support for emotion/sentiment conversion plugins.
Senpy includes support for emotion and sentiment conversion.
When a user requests a specific model, senpy will choose a strategy to convert the model that the service usually outputs and the model requested by the user.
Out of the box, senpy can convert from the `emotionml:pad` (pleasure-arousal-dominance) dimensional model to `emoml:big6` (Ekman's big-6) categories, and vice versa.
This specific conversion uses a series of dimensional centroids (`emotionml:pad`) for each emotion category (`emotionml:big6`).
A dimensional value is converted to a category by looking for the nearest centroid.
The centroids are calculated according to this article:
.. code-block:: text
Kim, S. M., Valitutti, A., & Calvo, R. A. (2010, June).
Evaluation of unsupervised emotion models to textual affect recognition.
In Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text (pp. 62-70).
Association for Computational Linguistics.
It is possible to add new conversion strategies by `Developing a conversion plugin`_.
Use
===
Consider the original query: http://127.0.0.1:5000/api/?i=hello&algo=emotion-random
Consider the following query to an emotion service: http://senpy.gsi.upm.es/api/emotion-anew?i=good
The requested plugin (emotion-random) returns emotions using Ekman's model (or big6 in EmotionML):
The requested plugin (emotion-random) returns emotions using the VAD space (FSRE dimensions in EmotionML):
.. code:: json
... rest of the document ...
{
"@type": "emotionSet",
"onyx:hasEmotion": {
"@type": "emotion",
"onyx:hasEmotionCategory": "emoml:big6anger"
},
"prov:wasGeneratedBy": "plugins/emotion-random_0.1"
}
[
{
"@type": "EmotionSet",
"onyx:hasEmotion": [
{
"@type": "Emotion",
"emoml:pad-dimensions_arousal": 5.43,
"emoml:pad-dimensions_dominance": 6.41,
"emoml:pad-dimensions_pleasure": 7.47,
"prov:wasGeneratedBy": "prefix:Analysis_1562744784.8789825"
}
],
"prov:wasGeneratedBy": "prefix:Analysis_1562744784.8789825"
}
]
To get these emotions in VAD space (FSRE dimensions in EmotionML), we'd do this:
To get the equivalent of these emotions in Ekman's categories (i.e., Ekman's Big 6 in EmotionML), we'd do this:
http://127.0.0.1:5000/api/?i=hello&algo=emotion-random&emotionModel=emoml:fsre-dimensions
http://senpy.gsi.upm.es/api/emotion-anew?i=good&emotion-model=emoml:big6
This call, provided there is a valid conversion plugin from Ekman's to VAD, would return something like this:
.. code:: json
... rest of the document ...
[
{
"@type": "EmotionSet",
"onyx:hasEmotion": [
{
"@type": "emotionSet",
"onyx:hasEmotion": {
"@type": "emotion",
"onyx:hasEmotionCategory": "emoml:big6anger"
},
"prov:wasGeneratedBy": "plugins/emotion-random.1"
}, {
"@type": "emotionSet",
"onyx:hasEmotion": {
"@type": "emotion",
"A": 7.22,
"D": 6.28,
"V": 8.6
},
"prov:wasGeneratedBy": "plugins/Ekman2VAD_0.1"
"@type": "Emotion",
"onyx:algorithmConfidence": 4.4979,
"onyx:hasEmotionCategory": "emoml:big6happiness"
}
],
"prov:wasDerivedFrom": {
"@id": "Emotions0",
"@type": "EmotionSet",
"onyx:hasEmotion": [
{
"@id": "Emotion0",
"@type": "Emotion",
"emoml:pad-dimensions_arousal": 5.43,
"emoml:pad-dimensions_dominance": 6.41,
"emoml:pad-dimensions_pleasure": 7.47,
"prov:wasGeneratedBy": "prefix:Analysis_1562745220.1553965"
}
],
"prov:wasGeneratedBy": "prefix:Analysis_1562745220.1553965"
},
"prov:wasGeneratedBy": "prefix:Analysis_1562745220.1570725"
}
]
That is called a *full* response, as it simply adds the converted emotion alongside.
It is also possible to get the original emotion nested within the new converted emotion, using the `conversion=nested` parameter:
http://senpy.gsi.upm.es/api/emotion-anew?i=good&emotion-model=emoml:big6&conversion=nested
.. code:: json
[
{
"@type": "EmotionSet",
"onyx:hasEmotion": [
{
"@type": "Emotion",
"onyx:algorithmConfidence": 4.4979,
"onyx:hasEmotionCategory": "emoml:big6happiness"
}
],
"prov:wasDerivedFrom": {
"@id": "Emotions0",
"@type": "EmotionSet",
"onyx:hasEmotion": [
{
"@id": "Emotion0",
"@type": "Emotion",
"emoml:pad-dimensions_arousal": 5.43,
"emoml:pad-dimensions_dominance": 6.41,
"emoml:pad-dimensions_pleasure": 7.47,
"prov:wasGeneratedBy": "prefix:Analysis_1562744962.896306"
}
],
"prov:wasGeneratedBy": "prefix:Analysis_1562744962.896306"
},
"prov:wasGeneratedBy": "prefix:Analysis_1562744962.8978968"
}
]
... rest of the document ...
{
"@type": "emotionSet",
"onyx:hasEmotion": {
"@type": "emotion",
"onyx:hasEmotionCategory": "emoml:big6anger"
},
"prov:wasGeneratedBy": "plugins/emotion-random.1"
"onyx:wasDerivedFrom": {
"@type": "emotionSet",
"onyx:hasEmotion": {
"@type": "emotion",
"A": 7.22,
"D": 6.28,
"V": 8.6
},
"prov:wasGeneratedBy": "plugins/Ekman2VAD_0.1"
}
}
Lastly, `conversion=filtered` would only return the converted emotions.
.. code:: json
[
{
"@type": "EmotionSet",
"onyx:hasEmotion": [
{
"@type": "Emotion",
"onyx:algorithmConfidence": 4.4979,
"onyx:hasEmotionCategory": "emoml:big6happiness"
}
],
"prov:wasGeneratedBy": "prefix:Analysis_1562744925.7322266"
}
]
Developing a conversion plugin
================================
==============================
Conversion plugins are discovered by the server just like any other plugin.
The difference is the slightly different API, and the need to specify the `source` and `target` of the conversion.
......@@ -106,7 +165,6 @@ For instance, an emotion conversion plugin needs the following:
.. code:: python
......@@ -114,3 +172,6 @@ For instance, an emotion conversion plugin needs the following:
def convert(self, emotionSet, fromModel, toModel, params):
pass
More implementation details are shown in the `centroids plugin <https://github.com/gsi-upm/senpy/blob/master/senpy/plugins/postprocessing/emotion/centroids.py>`_.
......@@ -2,7 +2,7 @@ Demo
----
There is a demo available on http://senpy.gsi.upm.es/, where you can test a live instance of Senpy, with several open source plugins.
You can use the playground (a web interface) or make HTTP requests to the service API.
You can use the playground (a web interface) or the HTTP API.
.. image:: playground-0.20.png
:target: http://senpy.gsi.upm.es
......
......@@ -19,6 +19,7 @@ Sharing your sentiment analysis with the world has never been easier!
.. toctree::
:maxdepth: 1
server-cli
plugins-quickstart
plugins-faq
plugins-definition
......@@ -12,24 +12,97 @@ Welcome to Senpy's documentation!
.. image:: https://img.shields.io/pypi/l/requests.svg
:target: https://lab.gsi.upm.es/senpy/senpy/
Senpy is a framework to build sentiment and emotion analysis services.
It provides functionalities for:
Senpy is a framework for sentiment and emotion analysis services.
Senpy services are interchangeable and easy to use because they share a common semantic :doc:`apischema`.
- developing sentiment and emotion classifier and exposing them as an HTTP service
- requesting sentiment and emotion analysis from different providers (i.e. Vader, Sentimet140, ...) using the same interface (:doc:`apischema`). In this way, applications do not depend on the API offered for these services.
- combining services that use different sentiment model (e.g. polarity between [-1, 1] or [0,1] or emotion models (e.g. Ekkman or VAD)
- evaluating sentiment algorithms with well known datasets
If you interested in consuming Senpy services, read :doc:`Quickstart`.
Using senpy services is as simple as sending an HTTP request with your favourite tool or library.
Let's analyze the sentiment of the text "Senpy is awesome".
We can call the `Sentiment140 <http://www.sentiment140.com/>`_ service with an HTTP request using curl:
.. code:: shell
:emphasize-lines: 14,18
$ curl "http://senpy.gsi.upm.es/api/sentiment140" \
--data-urlencode "input=Senpy is awesome"
{
"@context": "http://senpy.gsi.upm.es/api/contexts/YXBpL3NlbnRpbWVudDE0MD8j",
"@type": "Results",
"entries": [
{
"@id": "prefix:",
"@type": "Entry",
"marl:hasOpinion": [
{
"@type": "Sentiment",
"marl:hasPolarity": "marl:Positive",
"prov:wasGeneratedBy": "prefix:Analysis_1554389334.6431913"
}
],
"nif:isString": "Senpy is awesome",
"onyx:hasEmotionSet": []
}
]
}
Congratulations, you’ve used your first senpy service!
You can observe the result: the polarity is positive (marl:Positive). The reason of this prefix is that Senpy follows a linked data approach.
You can analyze the same sentence using a different sentiment service (e.g. Vader) and requesting a different format (e.g. turtle):
.. code:: shell
$ curl "http://senpy.gsi.upm.es/api/sentiment-vader" \
--data-urlencode "input=Senpy is awesome" \
--data-urlencode "outformat=turtle"
@prefix : <http://www.gsi.upm.es/onto/senpy/ns#> .
@prefix endpoint: <http://senpy.gsi.upm.es/api/> .
@prefix marl: <http://www.gsi.dit.upm.es/ontologies/marl/ns#> .
@prefix nif: <http://persistence.uni-leipzig.org/nlp2rdf/ontologies/nif-core#> .
@prefix prefix: <http://senpy.invalid/> .
@prefix prov: <http://www.w3.org/ns/prov#> .
@prefix senpy: <http://www.gsi.upm.es/onto/senpy/ns#> .
prefix: a senpy:Entry ;
nif:isString "Senpy is awesome" ;
marl:hasOpinion [ a senpy:Sentiment ;
marl:hasPolarity "marl:Positive" ;
marl:polarityValue 6.72e-01 ;
prov:wasGeneratedBy prefix:Analysis_1562668175.9808676 ] .
[] a senpy:Results ;
prov:used prefix: .
As you see, Vader returns also the polarity value (0.67) in addition to the category (positive).
If you are interested in consuming Senpy services, read :doc:`Quickstart`.
To get familiar with the concepts behind Senpy, and what it can offer for service developers, check out :doc:`development`.
:doc:`apischema` contains information about the semantic models and vocabularies used by Senpy.
.. toctree::
:caption: Learn more about senpy:
:maxdepth: 2
:hidden:
senpy
demo
Quickstart.ipynb
installation
conversion
Evaluation.ipynb
apischema
advanced
development
publications
projects
......@@ -17,18 +17,20 @@ Through PIP
.. code:: bash
pip install --user senpy
pip install senpy
# Or with --user if you get permission errors:
pip install --user senpy
Alternatively, you can use the development version:
.. code:: bash
..
Alternatively, you can use the development version:
git clone git@github.com:gsi-upm/senpy
cd senpy
pip install --user .
.. code:: bash
If you want to install senpy globally, use sudo instead of the ``--user`` flag.
git clone git@github.com:gsi-upm/senpy
cd senpy
pip install --user .
Docker Image
************
......
......@@ -9,20 +9,21 @@ Lastly, it is also possible to add new plugins programmatically.
.. contents:: :local:
What is a plugin?
=================
..
What is a plugin?
=================
A plugin is a program that, given a text, will add annotations to it.
In practice, a plugin consists of at least two files:
A plugin is a program that, given a text, will add annotations to it.
In practice, a plugin consists of at least two files:
- Definition file: a `.senpy` file that describes the plugin (e.g. what input parameters it accepts, what emotion model it uses).
- Python module: the actual code that will add annotations to each input.
- Definition file: a `.senpy` file that describes the plugin (e.g. what input parameters it accepts, what emotion model it uses).
- Python module: the actual code that will add annotations to each input.
This separation allows us to deploy plugins that use the same code but employ different parameters.
For instance, one could use the same classifier and processing in several plugins, but train with different datasets.
This scenario is particularly useful for evaluation purposes.
This separation allows us to deploy plugins that use the same code but employ different parameters.
For instance, one could use the same classifier and processing in several plugins, but train with different datasets.
This scenario is particularly useful for evaluation purposes.
The only limitation is that the name of each plugin needs to be unique.
The only limitation is that the name of each plugin needs to be unique.
Definition files
================
......@@ -109,5 +110,3 @@ Now, in a file named ``helloworld.py``:
sentiment['marl:hasPolarity'] = 'marl:Negative'
entry.sentiments.append(sentiment)
yield entry
The complete code of the example plugin is available `here <https://lab.gsi.upm.es/senpy/plugin-prueba>`__.
......@@ -23,7 +23,7 @@ In practice, this is what a plugin looks like, tests included:
.. literalinclude:: ../example-plugins/rand_plugin.py
:emphasize-lines: 5-11