The transversality of Artificial Intelligence and its capability to deal with the challenging problems made AI techniques to pervade almost all sectors of the economy and society. Its ability to emulate intelligence provides a wide array of solutions such as supporting decision making, diagnosis generation, customer segmentation, predicting undesired situations, and more.
Due to its prominence in today’s sector, providing AI-based solutions is named the sexiest job of the century. According to Forbes, the global Machine Learning market is expected to reach $20.83B in 2024. But, there is a significant shortage of brilliant brains to meet the heavy demand in the AI sector. In this feature, we will discuss a few competitive and interesting Artificial Intelligence (AI) projects that can enhance your AI skills.
- Measuring distances between medical entities: DrugBank
- Towards a Universal Neural Network Encoder for Time Series
- Web Pattern Navigation Profiling for Online Marketing Campaigns
- Do Positive Twitter Messages Have an Impact on Stock Liquidity?
1. Measuring Distances Between Medical Entities: DrugBank
Processing medical data is a tough job for healthcare industries. They are facing a lot of difficulties in measuring medical entities -such as diseases, body parts, drugs, symptoms, and so on. This AI-based project presents an AI-based solution to measure the distance between the drugs based on a similar name, similar description, similar targets, and similar chemical compositions.
To identify the distance based on the above properties, we have to represent the drugs in a vector space model, and then we should analyze the textual similarity, semantic similarity, and chemical similarity.
Textual similarity aims to determine the level of similarity between two texts. Here, the text fields such as description, indication, and pharmacodynamics are concatenated to perform natural language processing. The NLP process includes methodologies such as removing stop words, transforming to lower case, and identifying the frequency-inverse document frequency (tf-idf). Then we use Latent Semantic Indexing (LSI) to reduce the dimension of the vector space model. And finally, the distance matrix is calculated using the Euclidean distance.
Next, we measure semantic similarity within a semantic space by using a knowledge base. Here, we calculate the distance between the corresponding synsets using WordNet. Then we compute the shortest path and the maximum depth of taxonomy using the Leacock and Chodorow metric.
Finally, we determine the chemical similarity, where we represent the molecule in the 2D form using fingerprint vectors and compute the similarity. The two fingerprint vectors used here are MACCS and ECFPS, and the coefficient of vectors are calculated using the Tanimoto Coefficient.
You can check the publisher’s version for more information.
2. Towards a Universal Neural Network Encoder for Time Series
Time series analysis has gathered much attention due to its quantitative and qualitative forecasting results. However, most approaches related to time series classifications focus on uniform length time-series data, one-dimensional data, and unlabelled data. But, in real-time cases, unequal time series data may arise because of various scenarios, which underlines the necessity to deal with a variable length of data. And also in certain cases such as demand forecasting and financial predictions, it is necessary to couple a global pattern with local calibration rather than relying on one-dimensional data. Above all, it is important to identify explainable features for time series data rather than providing unlabelled data.
In this Artificial Intelligence project, we use a universal neural network encoder to address the identified challenges related to variable length, high-dimension, and a few labeled data. The encoder we consider is a standard convolution network; Here, the convolution tackles the challenges related to high-dimension data. This convolution network is coupled with a time-wise attention mechanism to overcome the challenges related to variable-length data representations. And for labeled data, we use a universal network that is trained with a variety of datasets. Finally, to deal with time series classification COTE and HIVE-COTE algorithms are used.
You can check the publisher’s version for more information
3. Web Pattern Navigation Profiling for Online Marketing Campaigns
In today’s digital era, it is merely impossible for a majority of the population to perform their day to day activities without using the internet. This internet usage has opened up a lot of possibilities in product promotion and also in offering services. But with billions of people surfing the internet, it is important to target the right customer for your product promotions or to provide a solution based on their needs. The best strategy to understand the users’ behavior is by analyzing their online-navigation pattern. With Web-mining techniques, you can extract quality information related to the users’ interest and their needs.
This AI-based project aims to provide quality data through web pattern navigation profiling to tailor efficient online marketing campaigns. Here, we consider different navigation profiles based on the consecutive ordered sequence of domain visited, rather than considering a bag of websites. It examines certain socio-demographic profiles such as region, age, social class, and the number of relatives in the home, and identifies the frequent contiguous sequences.
The data used in this project represents Spanish internet users and is owned by Kantar Worldpanel. And for data processing, the M3 algorithm is used to identify all different sequences of websites based on the visitors in a given sample. This algorithm is efficient in extracting standard sequences of websites for a particular user segment. Then the extracted data is corrected using Bonferroni correction and False Discovery Rate correction. Hence, you can tailor an efficient marketing campaign to target a specific user segment.
You can check the publisher’s version for more information.
4. Do Positive Twitter Messages Have an Impact on Stock Liquidity?
An accurate stock prediction model plays a key role in business planning, but at the same time, it is a challenging task. By analyzing the historical prices using a knowledge-based system and machine learning, you can analyze the stock market cycle. However, recent studies show that social media can make an impact on stock liquidity.
For the past few years, social media has changed the way people interact with each other. The microblogging services provided by these platforms facilitate the rapid spread of information across the globe. In this AI-based project, we analyze how twitter messages can influence the sentiments and behavior of stockholders.
The volatility plays a significant role in the stock market. For measuring the volatility of the stock market, it is necessary to consider the liquidity of stock shares. Since the liquidity directly depends on a company’s functionality, the company-related information published on social media has a huge impact on twitter sentiments.
Here we consider two datasets for stock prediction. One dataset has the information related to stock price or volume of the stock exchange on a certain period. This dataset is necessary because speed and price are the two important features related to liquidity. It plays a significant role in trade decision making. The second dataset has the information extracted from the archives of Twitter. By using both the datasets, you can identify the correlations and casualties between different variables and segments of the issues. For analyzing the data regarding liquidity, you should calculate it using Amihuds measure. And to analyze the correlation, the average sentiment polarity per day is considered along with the illiquidity rate. Finally, a linear correlation is determined using Pearson’s correlation coefficient.
You can check the publisher’s version for more information.
Summary
Artificial Intelligence is considered as emerging and groundbreaking technology, which is indispensable for today’s economic sector and society. According to reports, there is a huge demand for AI talents in the market, and the AI solutions are white-hot at present. In this feature, we discussed the top 4 advanced project ideas that enhance your AI skills. By working on these project ideas, you can become one of the brilliant brains that provide quality AI solutions. Let’s wrap up with Geoffrey Hinton words,
“I have always been convinced that the only way to get artificial intelligence to work is to do the computation in a way similar to the human brain. That is the goal I have been pursuing. We are making progress, though we still have lots to learn about how the brain actually works.” – Geoffrey Hinton, Famous AI Scientist.
