Differential Privacy Term Frequency Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for learning the frequency of emojis and deep links on client devices compromise user privacy and do not effectively update dictionaries with trending terms, relying on centralized data that can identify individual users.
Innovation Solution
Implementing a differential privacy system that uses local sketches to aggregate data from multiple clients, ensuring that a server can learn term frequencies without identifying individual user data, using algorithms like ε-local differentially private count-median-sketch and Hadamard ε-local differentially private count-median-sketch to maintain user privacy and update term frequencies across devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If centralized data collection and analysis is used to learn term frequencies, then productivity in learning trending terms is improved, but user privacy is compromised
Solution Approach 1:
The patent introduces differential privacy as an intermediary mechanism between data collection and analysis. A privacy budget parameter ε controls the trade-off: smaller values provide stronger privacy guarantees while larger values improve term frequency learning accuracy. The system adds calibrated noise to individual term frequency counts before aggregation, preventing identification of individual users while maintaining population-level trends.
Solution Approach 2:
The system dynamically adjusts the privacy budget parameter ε and noise scaling factors to balance privacy protection and learning effectiveness. The noise magnitude is proportional to the sensitivity of the query and inversely proportional to ε, allowing flexible control over the privacy-accuracy trade-off based on application requirements.
2Measurement precision
If clear text data is collected from users to learn frequency, then measurement precision of term frequency is improved, but user privacy is compromised
Solution Approach 1:
Differential privacy acts as a mediator that preserves measurement precision at the population level while protecting individual privacy. The system collects precise term frequency data from many users, adds noise calibrated to the desired privacy level, and publishes aggregated results. Individual contributions remain private while collective patterns are accurately captured.
Solution Approach 2:
Instead of collecting and storing actual user data, the system collects noisy copies of term frequency counts. These copies contain sufficient information to learn population-level trends but lack the precision to identify individual users. The copying process with added noise transforms sensitive data into useful but anonymized statistics.
3Object-affected harmful factors
If local usage data only is used for ordering suggestions, then user privacy is preserved, but adaptability of content suggestions worsens
Solution Approach 1:
The patent segments the frequency learning process into local and global components. Local dictionaries on client devices capture individual user preferences and maintain privacy, while global servers aggregate noisy frequency data from many users to identify trending terms. The segmented approach allows both personalized local suggestions and adaptive global trend incorporation.
Solution Approach 2:
The system merges local dictionary data with globally aggregated trending term frequencies to create enhanced suggestion lists. Local usage patterns provide personalized relevance, while global trends ensure suggestions remain current and adaptable to changing user interests across the population.
4Adaptability or versatility
If comprehensive user data is collected for personalized suggestions, then adaptability of content suggestions is improved, but device complexity increases
Solution Approach 1:
The patent extracts only the essential frequency information needed for personalized suggestions from comprehensive user data. Instead of collecting and processing all user interactions, the system extracts term frequency counts from local dictionaries and sends only these aggregated statistics to servers. This extraction minimizes data transmission and processing complexity while maintaining suggestion quality.
Data Source
AI summary
Systems and methods are disclosed for generating term frequencies of known terms based on crowdsourced differentially private sketches of the known terms. An asset catalog can be updated with new frequency counts for known terms based on the crowdsourced differentially private sketches. Known terms can have a classification. A client device can maintain a privacy budget for each classification of known terms. Classifications can include emojis, deep links, locations, finance terms, and health terms, etc. A privacy budget ensures that a client does not transmit too much information to a term frequency server, thereby compromising the privacy of the client device.