System for opinion reconciliation
AssigneeYahoo! Inc.
InventorsSrujana Merugu, Phillip L. Bohannon, Ashwin Kumar Machanavajjhala, Pedro DeRose
Life of this patent
Estimating the term…
Abstract
A system is disclosed for reconciling opinions generated by agents with respect to one or more predicates. The disclosed system may use observed variables and a probabilistic model including latent parameters to estimate a truth score associated with each of the predicates. The truth score, as well as one or more of the latent parameters of the probabilistic model, may be estimated based on the observed variables. The truth score generated by the disclosed system may enable publishers to reliably represent the truth of a predicate to interested users.
Claims25
A system for reconciling opinions generated by one or more agents in relation to a predicate, comprising:
- a processor;
- and
- a memory coupled with the processor, the memory comprising:
- observed variables comprising:
- opinion scores, each opinion score being associated one of the opinions generated by the one or more agents in relation to the predicate;
- an agent id associated with each of the one or more agents;
- a predicate id associated with the predicate;
- an observed agent characteristic associated with each agent;
- and
- an observed predicate characteristic associated with each predicate;
- a probabilistic model comprising latent parameters, the latent parameters comprising:
- a truth score associated with the predicate;
- and
- an opinion predictiveness score associated with each of the opinions;
- and
- instructions that, when executed, cause the processor to:
- estimate the latent parameters based on the observed variables using a maximum likelihood technique;
- and
- output the truth score associated with the predicate.
A method for reconciling opinions generated by one or more agents in relation to a predicate, comprising:
- obtaining observed variables comprising:
- opinion scores, each opinion score being associated with one of the opinions;
- an agent id associated with each of the one or more agents;
- a predicate id associated with the predicate;
- observed agent characteristics associated with each agent;
- and
- observed predicate characteristics associated with each predicate;
- estimating latent parameters of a probabilistic model based on the observed variables using a maximum likelihood technique, where the latent parameters comprise:
- a truth score associated with the predicate;
- and
- an opinion predictiveness score associated with each of the opinions;
- latent agent characteristics;
- and
- latent predicate characteristics;
- and
- outputting the truth score.
A product comprising:
- a computer-readable medium;
- and
- programmable instructions stored on the computer-readable medium that, when executed, cause a processor in an opinion reconciliation system to:
- receive multiple opinions, where each opinion is generated by an agent in relation to a predicate;
- receive observed variables comprising an opinion score associated with each opinion;
- estimate latent parameters of a probabilistic model based on the observed variables and on defined relationships between the latent parameters and the observed variables, where the latent parameters comprise:
- a truth score associated with each predicate;
- latent agent characteristics;
- and
- latent predicate characteristics;
- and
- output the truth score associated with each predicate.
Dependencies are inferred from the claim text.