Resource navigation using neural networks
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for resource navigation using neural networks.
1 . A method performed by one or more computers, the method comprising:
receiving a query;
processing the query using a query encoder neural network to generate an embedding of the query;
identifying a current resource in a collection of resources; and
generating a response to the query, comprising:
at each of a plurality of navigation iterations:
identifying a plurality of outgoing links from the current resource as of the navigation iteration, each outgoing link being a link to a respective other resource in the collection of resources from the current resource;
processing content from the current resource using a resource encoder neural network to generate an embedding of the current resource;
generating a respective embedding of each outgoing link;
generating, using a navigation policy neural network, a respective score for each outgoing link from at least the embedding of the outgoing link, the embedding of the current resource, and the embedding of the query;
selecting one of the outgoing links based on the respective scores for the outgoing links; and
updating the current resource to be the respective other resource linked to by the outgoing link; and
after a last navigation interaction, generating a response that characterizes the current resource after the last navigation iteration.
2 . The method of claim 1 , wherein the collection of resources comprises one or more web sites, and wherein each resource in the collection represents at least a portion of a web page in the one or more web sites.
3 . The method of claim 2 , wherein, for at least one of the navigation iterations, the current resource as of the navigation iteration represents a portion of a particular web page, and the plurality of outgoing links from the current resource as of the navigation iteration include (i) a respective link to each of one or more other portions of the particular web pages and (ii) one or more hyperlinks to other web pages in the collection.
4 . The method of claim 2 , wherein the query is submitted through an Internet search engine and wherein (i) the one or more web sites have not been indexed by the Internet search engine or (ii) content of one or more of the web pages in the one or more web sites has changed since the one or more web sites have been indexed by the Internet search engine.
5 . The method of claim 2 , wherein the one or more web sites are inaccessible to a search engine crawler.
6 . The method of claim 1 , wherein the query is received from a client device and wherein the method further comprises:
providing, to the client device, the response to the query.
7 . The method of claim 1 , wherein the response to the query includes a link to the current resource after the last navigation iteration.
8 . The method of claim 1 , wherein the response to the query includes (i) content from the current resource after the last navigation iteration, (ii) a summary of the content from the current resource after the last navigation iteration, or (iii) both.
9 . The method of claim 1 , wherein the resource encoder neural network comprises a self-attention neural network that is configured to process text tokens from the resource to generate a respective token embedding of each of the text tokens.
10 . The method of claim 1 , wherein selecting one of the outgoing links based on the respective scores for the outgoing links comprises:
selecting an outgoing link having a highest respective score or generating, from the respective scores, a probability distribution that includes a respective probability for each outgoing link and sampling an outgoing link from the probability distribution.
11 . The method of claim 1 , further comprising:
after performing the last navigation iteration, determining that termination criteria for generating the response have been satisfied.
12 . The method of claim 11 , wherein determining that termination criteria have been satisfied comprises:
determining that a threshold number of navigation iterations have been performed.
13 . The method of claim 11 , wherein determining that termination criteria have been satisfied comprises:
determining that a similarity between the embedding of the query and the embedding of the current resource after the last navigation iteration satisfies a threshold.
14 . The method of claim 1 , wherein generating, using a navigation policy neural network, a respective score for each outgoing link from at least the embedding of the outgoing link, the embedding of the current resource, and the embedding of the query comprises:
processing a navigation input comprising the embedding of the current resource and the embedding of the query using one or more neural network layers to generate a combined embedding; and
for each outgoing link, determining the respective score for the outgoing link based on a similarity between the combined embedding and the embedding of the outgoing link.
15 . The method of claim 14 , wherein the navigation input further comprises respective embeddings of the current resources as of any preceding navigation iterations.
16 . The method of claim 1 , wherein generating a respective embedding of each outgoing link comprises:
generating the respective embedding from at least (i) content of the respective other resource linked to by the outgoing link, (ii) text in the current resource corresponding to the outgoing link, or (iii) both.
17 . The method of claim 16 , wherein, generating the respective embedding from at least (i) content of the respective other resource linked to by the outgoing link, (ii) text in the current resource corresponding to the outgoing link, or (iii) both comprises:
generating the respective embedding from at least (i) content of the respective other resource linked to by the outgoing link, (ii) text in the current resource corresponding to the outgoing link, or (iii) both, and (iv) one or more of:
a first indicator that indicates whether the respective other resource has already been selected as the current resource at a previous navigation iteration, or
a second indicator that indicates whether the respective other resource is another portion of a same web page as the current resource or is a different web page from the current resource.
18 . The method of claim 16 , wherein generating the respective embedding from at least (i) content of the respective other resource linked to by the outgoing link, (ii) text in the current resource corresponding to the outgoing link, or (iii) both comprises:
processing (i) the content of the respective other resource linked to by the outgoing link (ii) the text in the current resource corresponding to the outgoing link, or (iii) both using the resource encoder neural network.
19 . The method of claim 1 , wherein identifying a current resource in a collection of resources comprises:
identifying the current resource by applying an information retrieval technique to identify a resource in the collection that is relevant to the query.
20 . The method of claim 1 , further comprising:
identifying one or more other resources in the collection of resources; and
for each identified other resource, performing navigation iterations to identify a respective final other current resource, wherein the response characterizes one or more of the respective final other current resources for the identified other resources.
21 . A system comprising:
one or more processors; and
one or more storage devices storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
receiving a query;
processing the query using a query encoder neural network to generate an embedding of the query;
identifying a current resource in a collection of resources; and
generating a response to the query, comprising:
at each of a plurality of navigation iterations:
identifying a plurality of outgoing links from the current resource as of the navigation iteration, each outgoing link being a link to a respective other resource in the collection of resources from the current resource;
processing content from the current resource using a resource encoder neural network to generate an embedding of the current resource;
generating a respective embedding of each outgoing link;
generating, using a navigation policy neural network, a respective score for each outgoing link from at least the embedding of the outgoing link, the embedding of the current resource, and the embedding of the query;
selecting one of the outgoing links based on the respective scores for the outgoing links; and
updating the current resource to be the respective other resource linked to by the outgoing link; and
after a last navigation interaction, generating a response that characterizes the current resource after the last navigation iteration.
22 . One or more non-transitory computer-readable storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:
receiving a query;
processing the query using a query encoder neural network to generate an embedding of the query;
identifying a current resource in a collection of resources; and
generating a response to the query, comprising:
at each of a plurality of navigation iterations:
identifying a plurality of outgoing links from the current resource as of the navigation iteration, each outgoing link being a link to a respective other resource in the collection of resources from the current resource;
processing content from the current resource using a resource encoder neural network to generate an embedding of the current resource;
generating a respective embedding of each outgoing link;
generating, using a navigation policy neural network, a respective score for each outgoing link from at least the embedding of the outgoing link, the embedding of the current resource, and the embedding of the query;
selecting one of the outgoing links based on the respective scores for the outgoing links; and
updating the current resource to be the respective other resource linked to by the outgoing link; and
after a last navigation interaction, generating a response that characterizes the current resource after the last navigation iteration.