Generation of user-specific time-to-live values using machine learning
Techniques are disclosed relating to generating user-specific time-to-live (TTL) values using machine learning. In various embodiments, a server system maintains a cache data store that is operable to store data for a plurality of users of a web service. In response to a cache miss for a request from a first one of the plurality of users, the server system may generate a user-specific TTL value for the first user. In various embodiments, generating the user-specific TTL value may include using a machine learning model to generate a predicted future access pattern for the first user that indicates a distribution of time periods during which the first user is expected to access the web service and, based on the predicted future access pattern, determining the user-specific TTL value for the first user.
1 . A computer-implemented method, comprising operations including:
maintaining, by a server system, a cache data store operable to store data for a plurality of users associated with a web service;
generating, by the server system using a machine learning model, a user-specific time-to-live (TTL) value for data associated with a first user of the plurality of users based on one or more predicted future accesses of the web service by the first user, wherein said generating comprises:
applying user access information associated with the first user to the machine learning model to generate the one or more predicted future accesses of the web service by the first user, the one or more predicted future accesses indicating a future time period during which the first user is predicted to access the web service; and
replacing, by the server system, an initial TTL value previously generated for the data associated with the first user with the user-specific TTL value, wherein replacing the initial TTL value includes updating the cache data store by storing the user-specific TTL value along with the data associated with the first user in the cache data store.
2 . The computer-implemented method of claim 1 , wherein the initial TTL value and the user-specific TTL value are for the data associated with the first user, of the plurality of users, that is stored in the cache data store.
3 . The computer-implemented method of claim 1 , wherein the user access information is indicative of an access history, by the first user, of the web service over a previous time period.
4 . The computer-implemented method of claim 3 , wherein the user access information includes a set of digits corresponding to the access history, each index of the set of digits indicating whether the first user accessed the web service during a given unit of time over the previous time period.
5 . The computer-implemented method of claim 1 , wherein generating, by the server system, using the machine learning model, the user-specific TTL value for the first user comprises further operations including:
analyzing the user access information to identify a series of sequential units of time during which the first user is predicted to access the web service; and
determining the user-specific TTL value based on the identified series of sequential units of time.
6 . The computer-implemented method of claim 5 , wherein the user-specific TTL value is selected as a number that corresponds to a last unit of time on which the first user is predicted to access the web service.
7 . The computer-implemented method of claim 1 , wherein the user-specific TTL value is associated with a specific type of data maintained in the cache data store, and wherein the server system is configured to generate a plurality of user-specific TTL values associated with different types of data for the first user.
8 . A non-transitory, computer-readable medium having instructions stored thereon that are executable by a server system to perform operations comprising:
maintaining a cache data store operable to store data for a plurality of users associated with a web service; and
responsive to a request for data associated with a first user of the plurality of users, generating, using a machine learning model, a user-specific time-to-live (TTL) value for the data based on one or more predicted future accesses of the web service by the first user, wherein said generating comprises:
applying user access information associated with the first user to the machine learning model to generate the one or more predicted future accesses of the web service by the first user, the one or more predicted future accesses indicating a future time period during which the first user is predicted to access the web service; and
updating the cache data store by replacing an initial TTL value with the user-specific TTL value, wherein the user-specific TTL value is stored along with the data associated with the first user within the cache data store.
9 . The non-transitory, computer-readable medium of claim 8 , wherein the operations further include:
transmitting requested data specified in the request for data to a user device associated with the first user; and
storing the requested data in the cache data store.
10 . The non-transitory, computer-readable medium of claim 9 , wherein the operations further include:
prior to an expiration of the user-specific TTL value, receiving a subsequent request from the first user for the requested data; and
servicing the subsequent request by retrieving the requested data from the cache data store.
11 . The non-transitory, computer-readable medium of claim 9 , wherein the operations further include:
determining that the user-specific TTL value for the first user has expired; and
evicting the requested data from the cache data store based on the determination.
12 . The non-transitory, computer-readable medium of claim 8 , wherein the machine learning model is implemented using one or more recurrent neural networks (RNNs).
13 . The non-transitory, computer-readable medium of claim 8 , wherein the operations further include:
analyzing the user access information to identify a series of consecutive units of time during which the first user is predicted to access the web service; and
determining the user-specific TTL value based on the identified series of consecutive units of time.
14 . The non-transitory, computer-readable medium of claim 8 , wherein the user-specific TTL value and requested data specified in the request for data are associated with a first instrument type, and wherein the operations further include:
receiving another request from the first user for data associated with a second instrument type; and
generating, using the machine learning model, a separate user-specific TTL value for the second instrument type based on user access information associated with the second instrument type.
15 . A system comprising:
a processor; and
a non-transitory computer-readable medium having stored thereon instructions that are executable by the processor to cause the system to perform operations comprising:
maintaining a cache data store operable to store data for a plurality of users associated with a web service;
generating using a machine learning model, a user-specific time-to-live (TTL) value for data associated with a first user of the plurality of users based on one or more predicted future accesses of the web service by the first user, wherein said generating comprises:
applying user access information associated with the first user to the machine learning model to generate the one or more predicted future accesses of the web service by the first user, the one or more predicted future accesses indicating a future time period during which the first user is predicted to access the web service; and
replacing an initial TTL value previously generated for the data associated with the first user with the user-specific TTL value, wherein replacing the initial TTL value includes updating the cache data store by storing the user-specific TTL value along with the data associated with the first user in the cache data store.
16 . The system of claim 15 , wherein the initial TTL value and the user-specific TTL value are for the data associated with the first user, of the plurality of users, that is stored in the cache data store.
17 . The system of claim 15 , wherein the user access information is indicative of an access history, by the first user, of the web service over a previous time period.
18 . The system of claim 17 , wherein the user access information includes a set of digits corresponding to the access history, each index of the set of digits indicating whether the first user accessed the web service during a given unit of time over the previous time period.
19 . The system of claim 15 , wherein the instructions are further executable to cause the system to train the machine learning model by:
assessing an accuracy of an access history predicted by the machine learning model for the first user; and
calculating a cost function based on the assessment.
20 . The system of claim 15 , wherein the machine learning model is a long short-term memory (LSTM) artificial neural network.