{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":651694457,"defaultBranch":"main","name":"TikTok-ML-Analysis-on-Turkish-Political-Parties","ownerLogin":"UygarTalu","currentUserCanPush":false,"isFork":false,"isEmpty":false,"createdAt":"2023-06-09T20:49:49.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/113293853?v=4","public":true,"private":false,"isOrgOwned":false},"refInfo":{"name":"","listCacheKey":"v0:1690146502.0","currentOid":""},"activityList":{"items":[{"before":"41259b6ca5d100b0ac57c7d3d6ae419c4d6d80d4","after":"c2b31039567aed1b689b8623a449d6d772fc8b83","ref":"refs/heads/main","pushedAt":"2023-07-23T21:08:22.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nThis is the complete script, where you can find all the analysis layers with the code explanations inside.","shortMessageHtmlLink":"Add files via upload"}},{"before":"19776d364e8ecca9b2d49bcdde2cfc5fb64303cf","after":"41259b6ca5d100b0ac57c7d3d6ae419c4d6d80d4","ref":"refs/heads/main","pushedAt":"2023-07-23T21:07:25.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nThe objective of the final layer of machine learning analysis is to cluster videos based \r\non the specific features extracted or produced in earlier layers. The primary purpose \r\nof clustering in this instance is to comprehend the strategic maneuvers employed by \r\nthe analyzed political parties. The KMeans clustering algorithm, the Elbow method, \r\nand the StandardScaler method, all from the fields of machine learning and statistics, \r\nare the primary tools utilized in this layer to accomplish this objective.\r\n\r\nThis layer of analysis is of great importance, not only because it builds upon the robust \r\nand complex feature generation of previous layers, but also because it provides a \r\nconcrete foundation for understanding the relative positions of the parties in terms of \r\ntheir videos and associated features. The critical insights that emerge from this layer \r\ninclude the identification of key video characteristics that the parties use strategically \r\nto define their positions or differentiate themselves. The analysis and resulting clusters \r\nfacilitate the visualization of relationships and serve as a focal point for the \r\nidentification of strategic moves.\r\n\r\nThis layer presents, in summary, a methodical strategy for discerning the strategic \r\nmoves of political parties through the clustering of their videos based on their \r\ngenerated features. It employs sophisticated machine learning techniques, rigorous \r\noptimization procedures, and careful interpretation of results, all of which serve to \r\nimprove our understanding of these parties' strategic positioning. In addition, the \r\n25\r\nlayer emphasizes the importance of ensuring equal feature contribution in the \r\nclustering process, identifies the optimal number of clusters, and paves the way for \r\nstrategic insights to be gleaned from the videos of these political parties. This layer's \r\nmethodology and findings provide a solid foundation for future research and have \r\nsignificant potential for application in political strategy analysis.","shortMessageHtmlLink":"Add files via upload"}},{"before":"560c172adda95d14aafc3ef202a81c3ff7666f76","after":"19776d364e8ecca9b2d49bcdde2cfc5fb64303cf","ref":"refs/heads/main","pushedAt":"2023-07-23T21:06:44.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Delete CLUSTERING_ANALYSIS_ML_ANALYSIS_LAYER_7_THESIS.py","shortMessageHtmlLink":"Delete CLUSTERING_ANALYSIS_ML_ANALYSIS_LAYER_7_THESIS.py"}},{"before":"8921b887369dc5c3f4b8a951748ac061298685d2","after":"560c172adda95d14aafc3ef202a81c3ff7666f76","ref":"refs/heads/main","pushedAt":"2023-07-23T21:05:04.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nThe objective of the final layer of machine learning analysis is to cluster videos based \r\non the specific features extracted or produced in earlier layers. The primary purpose \r\nof clustering in this instance is to comprehend the strategic maneuvers employed by \r\nthe analyzed political parties. The KMeans clustering algorithm, the Elbow method, \r\nand the StandardScaler method, all from the fields of machine learning and statistics, \r\nare the primary tools utilized in this layer to accomplish this objective.\r\n\r\nThis layer of analysis is of great importance, not only because it builds upon the robust \r\nand complex feature generation of previous layers, but also because it provides a \r\nconcrete foundation for understanding the relative positions of the parties in terms of \r\ntheir videos and associated features. The critical insights that emerge from this layer \r\ninclude the identification of key video characteristics that the parties use strategically \r\nto define their positions or differentiate themselves. The analysis and resulting clusters \r\nfacilitate the visualization of relationships and serve as a focal point for the \r\nidentification of strategic moves.\r\nThis layer presents, in summary, a methodical strategy for discerning the strategic \r\nmoves of political parties through the clustering of their videos based on their \r\ngenerated features. It employs sophisticated machine learning techniques, rigorous \r\noptimization procedures, and careful interpretation of results, all of which serve to \r\nimprove our understanding of these parties' strategic positioning. In addition, the \r\n25\r\nlayer emphasizes the importance of ensuring equal feature contribution in the \r\nclustering process, identifies the optimal number of clusters, and paves the way for \r\nstrategic insights to be gleaned from the videos of these political parties. This layer's \r\nmethodology and findings provide a solid foundation for future research and have \r\nsignificant potential for application in political strategy analysis.","shortMessageHtmlLink":"Add files via upload"}},{"before":"d919b54707ac985169f3441996207962a17b41bf","after":"8921b887369dc5c3f4b8a951748ac061298685d2","ref":"refs/heads/main","pushedAt":"2023-07-23T21:04:06.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nFeature engineering bridges the gap between raw data and effective models. It involves creating \r\nnew features from existing ones in order to improve the learning process and \r\nperformance of machine learning models. In this TikTok video analysis project, \r\nfeature engineering is essential to the generation of insights that could help us better \r\ncomprehend the strategic maneuvers of two political parties. This layer of analysis is \r\ncomprehensive, sophisticated, and indispensable for enhancing the preceding machine \r\nlearning analysis layers.\r\n\r\nBelow you can find what does the features signal for, \r\n\r\nEngagement Rate: This is the first metric calculated, and it represents the ratio of likes \r\nto views. It measures the video's popularity or appeal. High engagement rates \r\n21\r\nindicate not only passive consumption, but also active participation from users with \r\nthe latter taking the additional step of approving the content as opposed to simply \r\nviewing it.\r\nThe 'Face_Detection_Rate_FE' metric measures the rate at which faces are detected in \r\na video. These two characteristics reveal the audience-resonating strategies employed \r\nby the parties. A high face detection rate may indicate that parties are employing \r\nrecognizable faces to gain support.\r\nFace Detection Rate: This metric represents the frequency of detected faces in a video, \r\na potentially significant factor when analyzing videos in which human presence or \r\nfacial expressions are relevant. Higher face detection rates may indicate a video's \r\nemphasis on human subjects or emotions, providing insight into the parties' emphasis \r\non individual politicians or narratives centered on humans.\r\nThe 'Dominant Emotion Score' identifies the most prominent emotion expressed in the \r\nvideo. The dominant emotion score, on the other hand, provides insight into the party's \r\ncommunication strategy by revealing the emotion they wish to convey. This \r\ninteraction contributes to a comprehensive understanding of the strategic positioning \r\nof the parties.\r\nThe Dominant Emotion Score identifies the dominant emotion in the video. It aids in \r\ncomprehending the emotional tone of the parties' social media campaigns, thereby \r\nshedding light on their communication strategies.\r\nWith the emotional diversity and sentiment disparity features, we explore the \r\nemotional landscape of the video content in greater depth. Our fourth feature, \r\n'Emotional_Diversity_FE', represents the variety of emotions expressed in the video, \r\nwhereas our fifth feature, 'Sentiment_Disparity_FE', captures the disparity in \r\nsentiment between the comments and the recognized speeches. We capture the \r\nemotional complexity of the content and its reception by comprehending the range of \r\nemotions expressed and the sentiment difference between the videos and the \r\ncomments. It offers a unique perspective on how effectively the party conveys its \r\nmessages and how the audience interprets them.\r\nDiversity of Emotions: This is a measurement of the variety of emotions displayed in \r\na video. Greater emotional diversity is indicative of a more complex emotional \r\nnarrative. This could help determine whether a party is focusing on eliciting a single \r\nemotion or employing a more diverse emotional strategy.\r\n22\r\nThis illustrates the disparity in tone between the comments and the recognized \r\nspeeches in the videos. A positive value indicates that the comments are more positive \r\nthan the speeches, while a negative value indicates the opposite. It enables the \r\npotential understanding of differences in sentiment perception between spoken content \r\nand textual comments.\r\nSixth on our list, the 'Engagement_Per_Second_FE' feature quantifies the average \r\nengagement per second. This feature can reveal whether the party's videos are \r\nengrossing enough to maintain the audience's attention throughout, or whether they \r\nlose their appeal over time.\r\nEngagement Per Second: This metric determines the average engagement per second \r\nfor a video. Higher values indicate higher levels of engagement or interaction during \r\na given time period. This could be useful for comparing the engagement levels of \r\nvideos with varying lengths or identifying segments with higher viewer engagement.\r\nThe 'Topic_Alignment_Score_FE' function measures the semantic alignment between \r\nthe topics in the videos and the comments. This feature reveals whether the audience \r\nis in sync with the party's discussions or if there is a disconnect. It is a key indicator \r\nof the party's appeal to its audience.","shortMessageHtmlLink":"Add files via upload"}},{"before":"896854fda1a8187c7645c4a338113588c738beee","after":"d919b54707ac985169f3441996207962a17b41bf","ref":"refs/heads/main","pushedAt":"2023-07-23T21:01:42.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nThis analytical layer mirrors the fourth analysis layer in utilizing machine learning-based assessments. However, it presents variations in parameter specifications employed for Natural Language Processing (NLP) on discerned speeches and comments. Additionally, it employs divergent methodologies and algorithms, further differentiating it from the preceding layer.","shortMessageHtmlLink":"Add files via upload"}},{"before":"e94db98fbcf020d246edb2da582a529819584642","after":"896854fda1a8187c7645c4a338113588c738beee","ref":"refs/heads/main","pushedAt":"2023-07-23T20:59:45.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nThe fourth level of the multi-layered approach of our machine learning model focuses \r\non applying natural language processing (NLP) to TikTok video transcripts and \r\nassociated comments. This layer delves into linguistic \r\nanalysis in order to comprehend the underlying topical structures and sentiments \r\nexpressed in the video content and associated discussions. This step is crucial to our \r\nresearch because it provides us with nuanced insights into the semantic composition \r\nof the data and enables us to understand user sentiment towards various topics, both \r\nof which can be utilized to better comprehend the sociopolitical dynamics of the \r\nplatform.","shortMessageHtmlLink":"Add files via upload"}},{"before":"81105af75955a3e34256d70ee6223404c86a06c6","after":"e94db98fbcf020d246edb2da582a529819584642","ref":"refs/heads/main","pushedAt":"2023-07-23T20:56:13.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nThe third layer of our multi-layered TikTok political network analysis \r\nintegrates speech recognition to identify the specific narratives and topics conveyed \r\nin the videos. By carefully analyzing the spoken content, we can gain a deeper \r\nunderstanding of the nuances of the ongoing political discussions. This layer is \r\nessential for the creation of additional features for each video, thereby enhancing our \r\nknowledge of the political leanings of the TikTok network.\r\nThe underlying assumption is that the subjects discussed by a content creator and the \r\nsentiments they express may significantly influence the prospective voters in the \r\nTikTok network. Deciphering the spoken content is therefore \r\nessential for determining the messages being transmitted, the topics of the discussions, \r\nand the nature of the emotions evoked.\r\n\r\nWe've designed two major functions for this layer of analysis that work in tandem. \r\nThe initial section is dependent on audio extraction from the videos. For this purpose, \r\nwe utilize MoviePy, a robust Python library for video editing that offers a variety of \r\nfeatures, including audio extraction from video files. The primary \r\nclass used for audio extraction is VideoFileClip, which provides access to the audio \r\ncomponent of a video file.","shortMessageHtmlLink":"Add files via upload"}},{"before":"9149f1cbfa2b798e7a2a78dd82c4a5ba3bbdc1f7","after":"81105af75955a3e34256d70ee6223404c86a06c6","ref":"refs/heads/main","pushedAt":"2023-07-23T20:54:00.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nIn the second layer of this multi-layered machine learning analysis, face detection and \r\nemotion recognition technologies are utilized. This phase's primary objective is to \r\nidentify the emotions expressed by content creators affiliated with the two polarized \r\npolitical parties in question. Rather than emphasizing the recognition of specific \r\npoliticians or party-related figures, the primary focus is on understanding the emotive \r\nmessages conveyed by the content, irrespective of the individuals featured.\r\n11\r\nVideo content associated to the political party CHP provides an illustration of this \r\nstrategy. A video may not feature a prominent politician, but instead feature a satisfied \r\ncitizen praising the CHP's initiatives. The happiness or contentment depicted in this \r\nvideo would then be interpreted as representative of the party's emotional tone, despite \r\nthe absence of recognizable political figures.","shortMessageHtmlLink":"Add files via upload"}},{"before":null,"after":"9149f1cbfa2b798e7a2a78dd82c4a5ba3bbdc1f7","ref":"refs/heads/main","pushedAt":"2023-07-22T20:59:19.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"UygarTalu","name":"Uygar","path":"/UygarTalu","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/113293853?s=80&v=4"},"commit":{"message":"Add files via upload\n\nThe Anomaly Detection layer played an integral role in the core of our multi-layered \r\nanalysis process. This layer laid the groundwork for our investigation into the \r\nfascinating world of TikTok in the context of the Turkish elections of 2023. The \r\napplication of machine learning techniques, specifically the Isolation Forest algorithm, \r\nwas central to our strategy, and the algorithm's effectiveness in detecting anomalies \r\nhelped shed light on the complexities of the TikTok landscape during this politically\u0002charged period.\r\n9\r\nThere were two reasons for incorporating the Isolation Forest algorithm. Our primary \r\nobjective was to maximize our computational resources. By focusing our analysis on \r\nthe most influential videos, we were able to significantly reduce the high \r\ncomputational cost associated with the application of facial and emotional recognition \r\ntechniques in subsequent stages of the analysis. The second objective was to develop \r\na comprehensive understanding of the engagement metrics on the TikTok platform, \r\nwhich we hoped to accomplish by identifying the 'impactful' videos or anomalies in \r\nour dataset.","shortMessageHtmlLink":"Add files via upload"}}],"hasNextPage":false,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAADW2SR-QA","startCursor":null,"endCursor":null}},"title":"Activity · UygarTalu/TikTok-ML-Analysis-on-Turkish-Political-Parties"}