Techniques for improving standardized data accuracy
Described herein is a technique for mapping the raw text of a job title of an online job posting to an entity embedding, associated with an entity or entry of a title taxonomy. The raw text of the job title is first encoded to generate a multilingual word embedding in a multilingual word embedding space. Then, the vector representation of the job title, as represented in the multilingual word embedding space is translated, using a neural network, to a vector representation of the job title in the entity embedding space. Finally, a nearest neighbor search is performed to identify an entity embedding associated with an entity or entry in the title taxonomy that has a vector representation that is closest in distance to the vector output by the neural network.
1. A computer-implemented method, comprising:
training a multilayer perceptron neural network using training data, wherein an instance of training data in the set of training data includes a first vector representation of a job title in a multilingual word embedding space and a second vector representation of an entity in an entity embedding space, wherein the job title corresponding with the first vector representation is a job title that corresponds with a job title associated with the entity represented by the second vector representation in the entity embedding space;
providing as input to an input layer of the multilayer perceptron neural network a vector representation of a job title of an online job posting, the vector representation of the job title of the online job posting derived by mapping one or more words identified in raw text of the job title to one or more pre-trained multilingual word embeddings in the multilingual word embedding space, wherein a pre-trained multilingual word embedding comprises a vector representation of the job title expressed in multiple languages;
with the multilayer perceptron neural network, processing the input to translate the vector representation of the job title of the online job posting in the multilingual word embedding space to a vector representation of the job title of the online job posting in the entity embedding space associated with a multilingual title taxonomy;
performing a nearest neighbor search to identify one or more vector representations corresponding with one or more entity embeddings in the entity embedding space, each of the one or more entity embeddings associated with a job title from the multilingual title taxonomy; and
storing with the online job posting at least one of the one or more vector representations corresponding with the entity embedding in the entity embedding space.
2. The computer-implemented method of claim 1 , further comprising:
prior to providing the vector representation of the job title of the online job posting to the input layer of the multilayer perceptron neural network, processing the raw text of the job title of the online job posting with an entity tagger model to identify the one or more words in the raw text of the job title as words that are representative of the job title.
3. The computer-implemented method of claim 2 , further comprising:
responsive to the entity tagger model identifying two or more words in the raw text of the job title as words that are representative of the job title, calculating the vector representation of the job title of the online job posting by deriving an average of the vector representations for each of the two or more words in the raw text of the job title that have been identified as representative of the job title.
4. The computer-implemented method of claim 1 , wherein performing the nearest neighbor search to identify one or more vector representations corresponding with one or more entity embeddings in the entity embedding space comprises performing an approximate nearest neighbor search.
5. The computer-implemented method of claim 1 , further comprising:
subsequent to performing the nearest neighbor search to identify the plurality of vector representations:
for each entity embedding corresponding with a vector representation in the plurality of vector representations, ranking the job title from the title taxonomy associated with the entity embedding; and
storing with the online job posting each vector representation of the plurality of vector representations corresponding with the entity embeddings in the entity embedding space with a ranking score for the job title associated with vector representation.
6. The computer-implemented method of claim 5 , wherein ranking each job title from the title taxonomy comprises:
deriving the ranking score for each job title from the multilingual title taxonomy by providing as input to a machine learned model input features associated with the job title of the job posting and information associated with an entity in the multilingual title taxonomy corresponding with the entity embedding identified via the nearest neighbor search.
7. The computer-implemented method of claim 1 , further comprising:
obtaining a set of training data for training the multilayer perceptron neural network.
8. A system comprising:
a processor; and
a memory storage device storing instructions thereon, which, when executed by the processor, cause the system to:
train a multilayer perceptron neural network using training data, wherein an instance of training data in the set of training data includes a first vector representation of a job title in a multilingual word embedding space and a second vector representation of an entity in an entity embedding space, wherein the job title corresponding with the first vector representation is a job title that corresponds with a job title associated with the entity represented by the second vector representation in the entity embedding space;
provide as input to an input layer of the multilayer perceptron neural network a vector representation of a job title of an online job posting, the vector representation of the job title of the online job posting derived by mapping one or more words identified in raw text of the job title to one or more pre-trained multilingual word embeddings in the multilingual word embedding space, wherein a pre-trained multilingual word embedding comprises a vector representation of the job title expressed in multiple languages;
with the multilayer perceptron neural network, process the input to translate the vector representation of the job title of the online job posting in the multilingual word embedding space to a vector representation of the job title of the online job posting in the entity embedding space associated with a multilingual title taxonomy;
perform a nearest neighbor search to identify one or more vector representations corresponding with one or more entity embeddings in the entity embedding space, each of the one or more entity embeddings associated with a job title from the multilingual title taxonomy; and
store with the online job posting at least one of the one or more vector representations corresponding with the entity embedding in the entity embedding space.
9. The system of claim 8 , wherein the instructions, when executed by the processor, further cause the system to:
process the raw text of the job title of the online job posting with an entity tagger model to identify the one or more words in the raw text of the job title as words that are representative of the job title, prior to providing the vector representation of the job title of the online job posting to the input layer of the multilayer perceptron neural network.
10. The system of claim 9 , wherein the instructions, when executed by the processor, further cause the system to:
calculate the vector representation of the job title of the online job posting by deriving an average of the vector representations for each of the two or more words in the raw text of the job title that have been identified as representative of the job title in response to the entity tagger model identifying two or more words in the raw text of the job title as words that are representative of the job title.
11. The system of claim 8 , wherein the instructions, when executed by the processor, further cause the system to:
perform the nearest neighbor search to identify one or more vector representations corresponding with one or more entity embeddings in the entity embedding space by performing an approximate nearest neighbor search.
12. The system of claim 8 , wherein the instructions, when executed by the processor, further cause the system to:
subsequent to performing the nearest neighbor search to identify the plurality of vector representations:
for each entity embedding corresponding with a vector representation in the plurality of vector representations, ranking the job title from the title taxonomy associated with the entity embedding; and
store with the online job posting each vector representation of the plurality of vector representations corresponding with the entity embeddings in the entity embedding space with a ranking score for the job title associated with vector representation.
13. The system of claim 8 , wherein the instructions, when executed by the processor, further cause the system to rank each job title from the title taxonomy by:
derive a ranking score for each job title from the title taxonomy by providing as input to a machine learned model input features associated with the job title of the job posting and information associated with an entity in the title taxonomy corresponding with the entity embedding identified via the nearest neighbor search.
14. The system of claim 8 , wherein the instructions, when executed by the processor, further cause the system to:
obtain a set of training data for training the multilayer perceptron neural network.