Social video speaks many languages, trends evolve differently in every region, and content consumption varies across cultures. Understanding global audiences starts with speaking their language—literally. While English is the most widely spoken language in the world, other markets are emerging rapidly. Without the right insights, brands, agencies, and media companies risk missing opportunities to engage with audiences across the globe. This is why Tubular’s engineers have been working hard behind the scenes to expand video categorization coverage to include 4 new languages: Spanish, French, German, and Portuguese.
Multilanguage insights open up new worlds for marketers and media creators to:
- Craft localized campaigns with targeted messaging and positioning by monitoring regional content trends, cultural nuances, and audience preferences.
- Identify influential creators in key markets who are already engaging key target audiences in their native language.
- Optimize content strategies across languages and regions to maximize engagement in every market.
To provide a more holistic view of the social landscape, Tubular’s machine learning models organize billions of videos into a structured taxonomy of categories and topics. And now—we’ve expanded video categorization coverage to a total of 5 languages.
With this enhancement, Tubular provides broader, more precise global coverage, categorizing videos in Spanish, French, German, and Portuguese, increasing total video category coverage by 66%. New languages provide stronger insights into what’s trending, how audiences engage, and what drives views around the world.
To better understand how new language categorization opens doors for marketers and media creators, we spoke with Tubular Data Scientist II, Mykyta Minenko, and Senior Data Scientist I, Elena Dmytriieva, about the future of social video categorization.
What was the philosophy behind expanding Tubular’s product suite to include new language video categorization?
Our philosophy is driven by both technological advancement and market demands.
Initially, our system was heavily English-centric due to limitations in Natural Language Processing (NLP) models and the complexity of data collection. Expanding beyond English became a natural next step as NLP capabilities evolved—particularly with large language models (LLMS) demonstrating strong multilingual performance.
We selected the four new languages based on clients’ needs and locations. One of our biggest challenges in this expansion effort was data validation, given the limited labeled datasets for non-English videos. We addressed this by leveraging manual labeling, translations, and approximations to ensure accuracy. Looking ahead, our system is designed to scale, allowing for future expansions based on user demand.
Can you explain what happens behind the scenes when a user searches with the Video Categories filter?
Video categorization is not a real-time process. When new videos appear on social media, we first detect and collect them into our system. These videos are then processed in batches, where our models analyze and assign categories. When a user searches with the Video Categories filter, the system retrieves these precomputed category assignments from our database and applies the requested filters. This process remains consistent across all supported languages.
How does AI ensure accurate classification across different languages?
We use machine learning to embed video information into context vectors that capture the semantic meaning of the content. These vectors are then processed by our classifier, which assigns them to a structured taxonomy. This approach ensures accuracy and consistency, because AI excels at understanding nuances and context, leading to precise categorization.
Essentially, we’re leveraging AI’s multilingual capabilities to make sense of video content globally. With a 360° view of the social video ecosystem, marketers and media companies can easily understand shifts in content trends and audience behaviors.
What were the biggest challenges in expanding video categorization to multiple languages?
The primary challenge was the lack of training and validation data for non-English content. To overcome this, we used two key strategies:
- Multilingual LLMs: No matter what language is used in a video, the system organizes and categorizes it consistently, so that global media companies and brands can analyze video trends without language barriers.
- Transfer Learnings from English Data: Our classifier was initially trained on English-only examples. By applying it to non-English videos, we ensured classification quality remained high while minimizing distributional differences.
We also validate these assumptions to ensure accuracy, maintaining our high standards across all language offerings so that brands, agencies, and media creators can confidently identify the right trends and influencers, and optimize campaigns for global engagement without the risk of misleading data.
How did customer feedback influence this feature update?
Customer feedback is the driving factor and biggest point of inspiration for all Tubular product updates. It’s truly invaluable to us. We collect and prioritize customer feedback through our customer-facing teams, ensuring the most impactful insights are addressed. However, integrating feedback into video categorization is complex due to the number of use cases and touchpoints involved. To handle this, we have several strategies working at once.
First, we have a dedicated metadata operations team that refines our taxonomy based on industry trends and client need. Then, we rely on ongoing quality monitoring and validation processes that refine our categorization accuracy. And lastly, we have a system of manual overrides for fine-tuning model assignments when needed.
Looking ahead, how does Tubular plan to expand the multi-language categorization feature?
This is just the beginning! Our next major goal is to refine topics and keywords using large language models. Unlike categories, which serve as broad filters, topics and keywords provide deeper insights, allowing clients to:
- Identify emerging trends.
- Track specific creators and influencers.
- Cluster related videos to understand audience preferences.
- Analyze themes and conversations in greater detail.
Imagine using these multi-language patterns to tap into emerging markets and expand your reach—that’s where we’re headed!
Our vision is to provide the most comprehensive global social video intelligence, empowering our users to harness the full potential of social video across the world’s most widely spoken languages. Stay tuned for more exciting updates!
Interested in exploring our new video categories languages? Reach out to your Customer Success Manager or request a demo here to get started.