1. Analysis of the emotional coloring of messages on social networks and the search for signs of depression
According to the World Health Organization, depression is a serious problem that needs an urgent solution. More than 264 million people worldwide suffer from depression. Depression is the leading cause of disability in the world and contributes significantly to the global burden of disease. More than 800,000 people die from suicide every year due to depression. It is the second leading cause of death for people aged 15-29. Treatment for depression often begins later than necessary, treatment may be based on an inaccurate diagnosis, and sometimes depression is not treated at all.
The fact that the Internet has firmly entered the life of a modern person gives society a unique chance to detect early signs of depression. This is especially true of finding similar signs among young people. If we talk only about Twitter, it turns out that every second the users of this social network publish about 6,000 tweets. This means that about 350,000 tweets are published per minute, about 500 million per day, and about 200 billion per year.
According toPew Research Center About 72% of adults who use the Internet are social media users. Datasets taken from social media are important in many areas of research. For example - in the field of human sciences and medical research. Supporting such research through social media data analysis is rudimentary these days, and existing methods for analyzing such data are ineffective.
By analyzing linguistic markers in social media posts, it is possible to create a deep learning model that can detect the signs of depression in a particular netizen earlier than traditional methods.
Here are some related materials:
- You are what you tweet .
- : .
- .
2.
The idea behind this project is to generate accurate text summaries from video recordings of sports matches. There are sites that specialize in providing users with information about matches. Various models have been proposed aimed at extracting information about matches from video recordings and presenting it in text form. Neural networks are the best at this task. "Formation of text summaries" usually means the presentation of information in a concise form, with special attention to what carries facts and important information about the event.
To solve the problem of automatically creating a description of games from records, it is necessary to make sure that the models solving this problem could recognize especially important and exciting moments of games.
This can be achieved using some deep learning techniques such as 3D convolutional neural networks (3D-CNN), recurrent neural networks (RNN), long short term memory networks ( LTSM ). Other machine learning algorithms such as support vector machines (SVM) and k-means are also used here. In the course of applying such algorithms, the video is divided into parts, which are processed using the corresponding models.
Here is an article on the classification of sports video scenes for the purpose of generating summaries of them using transfer learning technology.
3. A system for solving handwritten equations based on convolutional neural networks
Recognizing handwritten math expressions is one of the challenging challenges facing machine vision researchers. Using convolutional neural networks ( CNNs ) and some image processing techniques can create a system that can recognize a handwritten mathematical expression . The development of such a system involves training the network using appropriately prepared datasets, represented by handwritten mathematical symbols.
Here are some resources on this topic:
- A system for solving handwritten equations based on a convolutional neural network .
- An example of an implementation of a CNN system for solving handwritten equations .
- Machine vision: Automatically check answer sheets for math assignments .
- Convert handwritten equations to LaTeX format .
4. Formation of brief reports on the materials of business meetings using natural language processing technologies
Have you ever found yourself in a situation where a long material needs to be reduced to a short synopsis? I had to deal with this during my studies. Namely, I had to spend a lot of time preparing some long essay, and the teacher had time only to read his brief annotation, which also took time to prepare.
The mechanisms for preparing brief summaries on some materials arose as an attempt to solve the problem of information overload, which a modern person is subject to. A system for extracting the most valuable information, for example, from the recording of certain negotiations or lectures, can be of great commercial and educational value. The development of such a system can be approached by applying a comprehensive analysis of textual information relevant to dialogues and monologues.
It takes a lot of time to manually create a summary of a report. But this problem can be solved using natural language processing ( NLP ) technologies .
To prepare a short annotation of the text, you can use the mechanisms based on deep learning that can "understand" the context of the entire text. Many would be just happy if they had a system at their disposal that could quickly and efficiently solve such problems.
Here are some articles about it:
- A comprehensive guide to developing a Python text abstracting system using deep learning technologies .
- The way to understand the essence of text summarization and development of your own summarization system in Python.
5. Implementation of a system that recognizes users' faces, determines their mood and offers them the appropriate music
A person's face reflects his inner state, from the face you can understand what emotions a person is experiencing. This information, for example, can be based on an automatic music selection system. The fact is that what kind of music people listen to often depends on their mood. Therefore, it is quite logical to assume that a system capable of “understanding” a person's mood and selecting suitable music for him has a future. Machine vision technologies can help us in solving this problem. They, in the recognition of emotions, involve the analysis of photographs or video clips.
APIs have already been created for solving such problems, which I find interesting and useful, although I have not had a chance to work with them yet. Here is the material about such APIs.
6. Search for habitable exoplanets based on images captured by space devices such as the Kepler telescope
In the last decade, a huge number of stars have been investigated for the presence of habitable planets around them. Manual data analysis to identify exoplanets is extremely time consuming and prone to human error. Convolutional neural networks are well suited for solving the problem of finding such planets.
- Here's an article on finding exoplanets using machine learning technologies.
- Here is a NASA press release on the use of artificial intelligence technologies in exoplanet searches.
7. Recover old damaged photos
Restoring old photos is hard work. This work can be facilitated by taking advantage of deep learning technologies. The corresponding system can automatically detect damage to images (kinks, scuffs, holes) and, using image reconstruction algorithms (Inpainting), get rid of the damage, restoring the lost parts of the photos.
Here are the related materials:
- Coloring and restoring old photos using deep learning techniques .
- Image Reconstruction Guide: Using Machine Learning Technologies to Eliminate Photo Defects .
- How to recover photos using machine learning and still do without tutorial datasets .
8. Making Music Using Deep Learning Technologies
Music is a collection of sounds of different frequencies. With this in mind, automatic music creation can be described as the process of creating small pieces of music with minimal human intervention. Machine learning professionals are at the forefront of computer music production technology these days.
Here are a couple of helpful materials on this:
Outcome
We've looked at eight promising ideas that can form the basis of projects that can enrich the portfolio of projects for the AI and machine learning practitioner. We hope you found something among these ideas that inspired you.
Are you planning to implement any of the above ideas?
