Вы находитесь на странице: 1из 6

EVOS SOLUTIONS

Simplifying Business

AI & ML Portfolio

Registered Office:
Evos Solutions Private Limited.
Plot No. - A/318, Saheed Nagar, Bhubaneswar, Odisha - 751007, India.

Phone: +91 94370 52454 E-mail: info@evossolutions.com Website: www.evossolutions.com


Introducing AI Business Line at Evos Solutions

Established in 2006, we are a specialized and energetic team offering technical solutions and services
across diverse industries. Innovation is at the core of our approach towards solving and simplifying
business challenges of our partners. Since 2017, our organization’s primary focus is deep into
Artificial Intelligence (AI) & Machine Learning (ML). Our team has a rich and diverse experience in
these domains. In near future we are also going to launch our first NLP (Natural Language
Processing) based product, Social Dobby. It is basically AI for managers.

Offerings in AI & ML

With our core expertise is machine learning, we step forward with high quality services at cost-
effective prices as per the suitability and business needs of our clients. Natural Language Processing
(NLP), Evolutionary Computation, Artificial Neural Networks (ANNs), Support Vector Machines
(SVM), and Deep Learning are some of the major domains that our team is competent with. In
relation to AI, we offer the following set of specialized services that are delivered through agile
software development process.

Sentiment Analysis Predictive Modelling

Prescriptive Modelling Recommender System

Tools & Frameworks in AI

TensorFlow Keras

Scikit-Learn spaCy

NLTK PyTorch

Weka R

Products

Social Dobby:
Project Social Dobby aims to build an aspect based sentiment analysis tool which helps its customers
(Product Owners, Brand Managers or Sales Managers) extract insights from social media data. The
users specify some aspect based keywords that have the maximum impact on their products and/or
services. The market sentiments extracted for these aspect based keywords enable the managers to
take better decisions strengthening their businesses.

2
Evos Solutions has built this NLP module which consists of a set of pipelines beginning from
extraction of social media data, pre-processing and prediction by a pre-trained machine learning
model delivering meaningful insights. The output is a JSON document which has the aspects and
their sentiments tagged and the same is consumed by the Social Dobby Dashboard Module for the
visualization layer.

AnaLITE:
AnaLITE, an android app, offers an on-device platform to perform predictions in any field that
includes finance to government schemes. Machine learning that facilitates systems to learn from
experience and perform analytics is the crux of the app that greatly supports the decision making
ability. Technically, it is powered by the machine learning algorithms represented in Weka and it is
designed to operate in less processing power, storage, and energy. AnaLITE app is fundamentally
composed of modules to handle all the important functionalities, such as building dataset, initializing
and processing of machine learning task, and managing results.

Projects
The following are the projects that the members of our team have experienced.

Aspect Based Sentiment Analysis:


Aspect Based Sentiment Analysis helps to identify and determine the sentiments towards specific
aspects in text. The problem consists of 2 sub-tasks, i.e., Aspect Term Extraction and Aspect Term
Polarity. In this project, Keras and Tensorflow have been used.

In Aspect Term Extraction, given a set of sentences with pre-identified entities (e.g., restaurants), we
identified the aspect terms present in the sentences. An aspect term names a particular aspect of
the target entity. To address this task Neural Attention Model (Deep Learning) has been used.

In Aspect Term Polarity, given one or more aspect terms within a sentence, determine whether the
polarity of each aspect term is positive, negative, neutral or conflict (i.e., both positive and negative).
Here, Recurrent Attention Network on Memory (Deep Learning) is used.

The models have been trained and tested on a dataset (Semeval) that has two domain-specific
datasets for laptops and restaurants, consisting of over 6,500 sentences where fine-grained aspect-
level human annotations is provided for training.

Resource: Novneet, Smriti Rekha Dey & Debasish Padhi (2019)

Stock Value Prediction:


Prediction of stock price has been beneficial for both the individual and institutional investors.
Precision of these forecast systems is necessary for better investment decisions with minimum risk
factors. To solve this problem, a hybrid model using Support Vector Machine (SVM) and Particle
Swarm Optimization (PSO) is developed. The performance of the model is evaluated with 18 years
daily transactional data (time series data) of Tata Steel stocks price from Bombay Stock Exchange
(BSE). It has been implemented as a standalone application in Java with Weka and JSwarm-PSO
(Version 2.08) which is an open source particle swarm optimization package.

Resource: Sabyasachi Mohanty (2018)

3
Topic Modelling:
Topics can be defined as “a repeating pattern of co-occurring terms in a corpus. It is an unsupervised
approach used for finding and observing the bunch of words (called “topics”) in large clusters of
texts. Here, LDA is used over Keras and Tensorflow framework.

Resource: Smriti Rekha Dey & Debasish Padhi (2018)

Movie Sentiment Analyzer:


The project computationally identify and categorize opinions related to movies. Two different
models are developed. One using LSTM and CNN (Deep learning neural network) and the other using
SVM (Machine Learning Model). The movie review dataset consists of 25,000 training data and
25,000 test data, and other 50,000 unlabelled data. The training set and the test set each contain
12,500 positive examples and 12,500 negative examples.

Resource: Smriti Rekha Dey (2018)

Prediction of Foodgrain Stock under Public Distribution System:


Forecasting of foodgrain stock under Public Distribution System can never be compromised by any
form of government in order to ensure food security. For this stacking ensemble hybrid model is
designed using Support Vector Regression (SVR), Teaching Learning Based Optimization (TLBO),
Kernel Principal Component Analysis (KPCA), and Elastic Net (EN). For this analytical study of
foodgrain stock, 99 features are selected. The model is implemented using java language with the
help of library files of LibSVM (Version 1.0.8), ElasticNet (Version 1.0.0) and Weka (Version 3.9.1)

Resource: Sabyasachi Mohanty (2018) (Part of PhD Work)

Prediction of Unskilled Labour Demand:


The prediction of unskilled labour demand is a multi-output prediction problem having 2 output
variables, i.e., Next month households demand and Next month persondays demand . Here, a hybrid model
of Multioutput Support Vector Regression (M-SVR) and Teaching Learning Based Optimization
(TLBO) is developed for the prediction of unskilled labour demand under Mahatma Gandhi National
Rural Employment Guarantee Act (MGNREGA), which is a scheme offered by Government of India to
enhance livelihood security of the rural poor through generation of wage employment
opportunities. The dataset consists of 18 features, 57 observations and 2 output variables. The
hybrid model has been implemented in Matlab with the help of library files of LibSVM.

Resource: Sabyasachi Mohanty (2018) (Part of PhD Work)

Haberman's Cancer Survival:


The project is all about analysing the data set and predicting whether the breast cancer patient will
survive after 5 years based upon different features and among all the features which is affecting
most.

Resource: Debasish Padhi (2018)

Amazon Fine Food Review Sentiment Analysis:


This project analyses the sentiments of reviews of different Amazon food products. This project is
focused to find the best model which can classify the class labels with high accuracy and less test
error. Here the source dataset consists of reviews of fine foods from amazon (kaggle). The python
based application has used Keras framework and have also used Numpy, Panda, Scikit-learn libraries.

4
Resource: Debasish Padhi (2018)

Personalized Cancer Diagnosis:


This project is focused on to determine cancer categories (malignant or benign) on the basis of genes
and the mutations. This classification problem has been addressed using Numpy, Panda, Scikit-learn.

Resource: Debasish Padhi (2018)

Analysis of Achievements of Government Schemes:


Forecasting physical achievements of welfare and developmental schemes offered by government
are always a challenging assignment. To predict the achievements of housing scheme (offered by
Government of India) a hybrid machine learning model based on Support Vector Regression, which
exhibits magnificent generalization capability on small samples with proper selection of hyper-
parameters using Teaching Learning Based Optimization along with Elastic Net for feature selection
is designed. For this a dataset pertaining to housing scheme of Government of India is used where
number of samples available is smaller than the features (76 features and 30 data points). It has
been implemented as a standalone application in Java with library files of LibSVM (Version 1.0.8),
elasticNet (Version 1.0.0) and Weka (Version 3.9.0).

Resource: Sabyasachi Mohanty (2017) (Part of PhD Work)

Movie Recommender:
It is a recommender system to recommend movies to the end users. It has implemented both
collaborative filtering model as well as content based filtering model. The models are trained using a
dataset having 1,00,000 ratings from 943 users on 1682 movies and demographic information of the
users (i.e., age, gender, occupation, etc.) are also taken into consideration.

Resource: Smriti Rekha Dey (2017)

GDP Prediction:
GDP is one of the generalized metrics for measuring the economic growth and activity of a country.
In order to predict GDP, a hybrid model is designed using Support Vector Regression (SVR), Teaching
Learning Based Optimization (TLBO) and Orthogonal Forward Selection (OFS). The model is trained
and tested on a dataset comprising of 36 features and 36 records. The hybrid model also
incorporated feature selection techniques. It has been implemented with Weka and LibSVM (Version
1.0.8).

Resource: Sabyasachi Mohanty (2016) (Part of PhD Work)

Prescriptive Model for GDP:


With an objective to maximize GDP, government tries to manage the available limited funds among
the necessities. Here the problem is to decide on budget allocations for Central Plan and for States
and Union Territories Plan with a goal to maximize GDP. For this, a hybrid model is designed using
Support Vector Regression (SVR), Teaching Learning Based Optimization (TLBO) and Orthogonal
Forward Selection (OFS). The model is trained and tested on a dataset comprising of 36 features and
36 records. It has been implemented as a standalone application in Java with Weka.

Resource: Sabyasachi Mohanty (2016) (Part of PhD Work)

5
Text Analytics Engine for Social Prism (a social listening platform):
An end-to-end pipeline for collecting data from web and social media, pre-processing it and running
ML algorithms for classification into pre-defined buckets.

The scope of the product includes a Text-analytics Engine useful in managing and analysis of
unstructured or semi-structured data, extraction of data like entities, sentiment, theme etc. from
web and social media and representation in the form of charts and graphs. From technology point of
view Machine Learning algorithms (both supervised and unsupervised), Java, Tomcat, Maven,
LIBSVM, LIB- LINEAR, Python libraries (NUMPY, SCIPY, SCIKIT, ANACONDA), R libraries have been
used.

The responsibilities includes

 Tweet classification as valid/invalid and subsequent categorization and sentiment analysis


using Machine Learning techniques, Lexicon-based sentiment analysis and Turney algorithm
 N-Gram based sentiment analysis - Unigrams, bigrams and trigrams included along with
discourse relationships as feature vectors
 Unsupervised clustering of tweets by LDA using Java based tool Mallet; helpful for
annotating huge number of tweets.
 Feature reduction of tweets for classification according to 80:20 principle using PCA.
 Webpage classification into sub-categories – Journalistic, Blog, Forum & Review using
Decision Tree
 Algorithm designing for domain classification using different algorithms such as Edit Distance
etc.
 Content extraction and meta-data (Title, Date, Comments) extraction for webpage
classification

Resource: Novneet (2014) (Project Assignment at CTS)

Resources
Sabyasachi Mohanty: An Engineering Graduate from ITER (Bhubaneswar), Master in Technology
from IIIT-Bhubaneswar in Computer Science & Engineering, and Doctor of Philosophy in Computer
Science & Engineering from Centurion University of Technology & Management.

Smriti Rekha Dey: An Engineering graduate from Orissa Engineering College in the field of Electrical
Engineering, and Master in Technology from IIIT-Bhubaneswar in Computer Science & Engineering.

Novneet: An Engineering graduate from IIT (Indian Institute of Technology), Hyderabad in Electrical
Engineering.

Debasish Padhi: An Engineering graduate from Centurion University of Technology & Management,
Bhubaneswar in Computer Science & Engineering.

Вам также может понравиться