The role of a social media manager in a digital agency gives the utmost importance to addressing important comments and messages on three main platforms: Twitter, Instagram, and Facebook. This can be done if a system of prioritization is put in place.
Sparrow MarketingWeb App
Industry: Marketing
Region: USA
SERCVICE USED
- Team Extension
- Web Development
- Mobile Deelopment
- UX/UI Design
Introduction
Challenge
The challenge lies in prioritizing comments and messages across social media according to their importance and creating a system where they can be interacted with accordingly.
Approach
We aimed to come up with a solution that is compatible as well as scalable in case the number of platforms involved increases. Initially, we gathered data (messages and comments) from all three platforms on a single platform and then developed a ranking algorithm based on parameter based classification and text based classification. Sentiment analysis was then used to categorize happy and angry comments.
Before initializing the project, we first break it down into the Backend Module. This module consists of several sub modules including data extraction module and API endpoints creation along with their subsequent testing as well. The most important part of this module was to extract data from platforms such as Twitter, Instagram and Facebook. Along with comments and messages, we also had to extract their passive (interactions with other pages) and active scores (interactions with our pages). These data points were then stored in DB through MYSQL connectors and then consequently fed to Machine Learning Algorithms. Finally, the priorities of comments and messages on all three platforms were stored back to DB. The endpoints using Swagger API were developed to bring data on the front end.
The main challenges came through the data extraction point of view as the rules and regulations about the data extraction of users have been quite strict and with limitations. We had to run a thorough research based on data extraction from Facebook (through Graph API), Twitter (through Twitter-Python) and Instagram (both official and non-official APIs). We were successful in extracting the data that we needed to train our model.
In our second model, the Machine Learning module, the target goal of Machine Learning was to categorize each message or comment by customers and then rank accordingly. This task was composed of two modules, that is, classification based on customers activity and classification based on content of comments/messages. Two classification models were used to train based on historical data. Both models will also be trained through inputs of social media managers while marking comments/messages as ignore, spam, skip or reply.
Outcome
By using the Backend and Machine Learning modules, applying parameters and several techniques, we were successful in prioritizing comments and messages to be addressed by social media managers. This allows the opportunity to better target and attract one’s audience with a focus on diversifying/streamlining the type of content to be created. It can also lay out the needs or building blocks required for an attainable plan and serve as a measure to track campaign results from.
Firstly, the parameter based classification has 5 parameters.
- Influence Score – Refers to user direct-activity score (likes on posts, comments on replies, reshares, pages liked, individual following, posts and activity) fetched from API data
- New Score – Users activity score based on minimum or no activity with the page.
- Happy Score – Happy sentiment score in the comment section or message sent by the user
- Angry Score – Angry Sentiment score in the comment section or message sent by the user.
- Embedding Score – Maximize probability of categorizing comments based on Word2vec embedding for comments and messages sent by users.
Secondly, text based classification (embedding scores) is based on training text data of comments and messages received from page, SMM response based on inputs from the user and the probability of class predicted serves as an embedding score to the main model, that is, parameter based classification.
Challenges Faced:
- Need at least 1000 data points (Messages/Comments) categorized as reply, skip, ignore, and spam, to set up a base text classification model for all three platforms.
- Need maximum active and passive features of customers that play part in priority response from the social media APIs for better accuracy.
Techniques used:
- Weighted sum and its normalization was used to map active and passive features to scores (0-1).
- Text data was preprocessed with special characters removal, spell checking, repetitive characters removal and tokenization.
- Sentiment Scores were produced using Nltk.Vader (used for social media sentiment text analysis).
- Word Embedding model was trained of 300 million twitter used to map text to features.
- SVM classifier was used as the classification model which was trained on the Word2vec feature vectors and final 5 Parameters.
Explore More View All
Sparrow Marketing
Prioritizing comments and messages for social media managers.
Industry: Marketing
Region: USA
Fine Grained Image Search
Semantic Image Search for Refined Results.
Industry: OCR
Region: USA
Real Time Food Searching & Recommendation
Matching restaurant food items to their respective recipes to gauge nutritional values to users.
Industry: Food
Region: USA
Braden Ericson | Founder & CEO Sparrow marketing
Datics team was great to work with. They were very proactive and helped me define the requirements for the project and were able to navigate the pivots I threw at them during our development period. They built a database, APIs, and an ML model beyond expectations. The team is actively fixing bugs that are popping up as we test and are a fun team to have weekly syncs with.
Email Us:
Let’s start discussing
