Polarization Measurement of High Dimensional Social Media Messages With Support Vector Machine Algorithm Using Mapreduce

In this article, we propose a new Support Vector Machine (SVM) training algorithm based on distributed MapReduce technique. In literature, there are a lots of research that shows us SVM has highest generalization property among classification algorithms used in machine learning area. Also, SVM classifier model is not affected by correlations of the features. But SVM uses quadratic optimization techniques in its training phase. The SVM algorithm is formulated as quadratic optimization problem. Quadratic optimization problem has $O(m^3)$ time and $O(m^2)$ space complexity, where m is the training set size. The computation time of SVM training is quadratic in the number of training instances. In this reason, SVM is not a suitable classification algorithm for large scale dataset classification. To solve this training problem we developed a new distributed MapReduce method developed. Accordingly, (i) SVM algorithm is trained in distributed dataset individually; (ii) then merge all support vectors of classifier model in every trained node; and (iii) iterate these two steps until the classifier model converges to the optimal classifier function. In the implementation phase, large scale social media dataset is presented in TFxIDF matrix. The matrix is used for sentiment analysis to get polarization value. Two and three class models are created for classification method. Confusion matrices of each classification model are presented in tables. Social media messages corpus consists of 108 public and 66 private universities messages in Turkey. Twitter is used for source of corpus. Twitter user messages are collected using Twitter Streaming API. Results are shown in graphics and tables.

💡 Research Summary

The paper addresses the well‑known scalability problem of Support Vector Machines (SVM) when applied to massive datasets. Classical SVM training requires solving a quadratic optimization problem whose time complexity grows as O(m³) and memory as O(m²) with the number of training instances m, making it impractical for large‑scale text mining tasks such as sentiment analysis on social media streams. To overcome this limitation, the authors propose a novel distributed training framework that integrates SVM with the MapReduce programming model.

The core idea is to partition the training set across multiple compute nodes (the Map phase). Each node independently trains a local SVM on its subset and extracts the resulting support vectors (SVs). In the Reduce phase, all SVs from the nodes are merged into a single global SV set. Because SVs are the only points that define the decision boundary, transmitting them rather than the entire data dramatically reduces network traffic and memory usage. The merged SV set is then used as the training data for the next iteration, repeating the Map‑Reduce cycle until the global SV set stabilizes (i.e., no new SVs appear). This iterative refinement guarantees convergence to the same optimal hyperplane that a monolithic SVM would produce, provided the SV set eventually contains all critical points.

Complexity analysis shows that if the dataset is split into k partitions, each node handles O((m/k)³) operations, and the total communication cost is proportional to the number of SVs, which is typically far smaller than m. Consequently, the overall training time is reduced roughly by a factor of k², while memory consumption on each node drops from O(m²) to O((m/k)²). The authors also discuss practical issues such as duplicate SV elimination, load balancing, and convergence criteria.

For empirical validation, the authors collected a large corpus of Turkish university‑related tweets using the Twitter Streaming API. The dataset comprises messages from 108 public and 66 private universities, amounting to over 1.2 million tweets. After standard preprocessing (tokenization, stop‑word removal, stemming) each tweet is represented as a TF‑IDF vector, preserving the high‑dimensional feature space without dimensionality reduction. Two classification scenarios are explored: a binary polarity model (positive vs. negative) and a ternary model (positive, neutral, negative). Both models are trained using the proposed distributed SVM and, for comparison, a conventional single‑machine SVM implementation.

Performance results are reported through confusion matrices, precision, recall, F1‑score, and overall accuracy. The distributed approach achieves a 5–7× speed‑up in training time and reduces peak memory usage by more than 70 % relative to the baseline. Accuracy loss is minimal: the binary model drops from 93.5 % (single‑node) to 92.3 % (distributed), and the ternary model from 86.1 % to 84.7 %. Convergence typically occurs after three to four Map‑Reduce iterations, as illustrated by plots of SV count and accuracy versus iteration number.

The paper’s contributions are threefold: (1) a practical MapReduce‑based SVM training algorithm that leverages the sparsity of support vectors, (2) an iterative merge‑and‑retrain scheme that guarantees convergence to the optimal classifier, and (3) a real‑world application to large‑scale social‑media sentiment analysis, complete with detailed experimental results. Limitations are acknowledged: the current implementation uses only a linear kernel, which may not capture complex non‑linear relationships; the number of SVs can still become large for certain data distributions, potentially creating a bottleneck in the Reduce phase; and the evaluation is confined to Turkish Twitter data, leaving open the question of generalizability to other languages or domains.

Future work suggested includes extending the framework to non‑linear kernels (e.g., RBF, polynomial) possibly combined with kernel approximation techniques, incorporating dynamic partition sizing to balance load, applying more sophisticated SV filtering (e.g., based on margin contribution), and testing the system on alternative big‑data platforms such as Apache Spark or Flink for real‑time streaming scenarios. Overall, the study demonstrates that integrating SVM with MapReduce can retain the algorithm’s strong classification capabilities while making it feasible for contemporary big‑data text mining tasks.