Project Background
In today's competitive market, brands and influencers are constantly searching for captivating content to enhance their marketing efforts. However, analyzing the target audience and account profile information can be a daunting task due to its complexity. Additionally, the sheer amount of data and rapidly evolving trends present a significant challenge for marketers. This innovative project utilizes AI technology to provide a solution to these obstacles, making it easier to find the most effective content.
Challenges & Requirements
To maintain a manageable scope for this project, it is imperative to incorporate a pre-trained word embedding model. Moreover, defining customized features and training the model with business data are essential steps to achieve success.
- The raw data is undergoing a revised classification process.
- Take into consideration only the features that are relevant.
- Reduce the data domain's range
- In order for the model to digest the data, it is necessary to extract it into TFrecord format.
- Constrain the prediction to only include hashtags, emojis, and mentions.
Code snippet showcasing data preparation:
Actions & Outcomes
- GloVe has been chosen as the pre-trained model, which is an unsupervised learning algorithm used to obtain vector representations for words.
- TensorFlow is utilized to orchestrate the training, testing, and prediction of the data model.
We have worked with Vincent on 2 projects now and on both occasions found him to be honest, hard working and professional. He has proven he has the skills and ability to complete work to a high standard, on time and on budget. I was personally very impressed with Vincents "can-do" attitude, nothing was too big or too small and even when we got to humps in the road, he always came back with a solution to overcome the problem. We have many more projects in the pipeline for 2012 and wouldn't hesitate to work with Vincent again. Thanks Vincent for all your hard work!!
Technologies Used
The model training and validation are executed using Jupyter notebook, while the model building logic and data preparation procedure are implemented using Python and TensorFlow. BigQuery and PostgresSQL are utilized to store classified data, while temporary files are stored in AWS S3.
Conclusions
As this project serves as a pivot, the prediction outcome may not be promising. Nevertheless, it lays the groundwork for future investigations into the potential benefits of deep learning in social media marketing and content creation.
