TECHNIQUES FOR DETERMINING SEGMENTS OF INFORMATION AREAS THROUGH ACTIVE ADAPTATION TO ENVIRONMENTAL CONTEXTS
The system optimizes data storage by promoting relevant data to higher layers for faster access and demoting less relevant data, addressing the challenges of large data storage in modern systems, improving efficiency and accuracy.
Patent Information
- Application Number
- DE112020002860
- Authority / Receiving Office
- DE · DE
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2019-08-07
- Filing Date
- 2020-07-21
- Publication Date
- 2024-11-28
- Estimated Expiration
- 2040-07-21
AI Technical Summary
Modern data processing systems face challenges with large data storage, including high costs, slow access times, and increased latency, which affect the efficiency and accuracy of information retrieval.
A system that actively promotes and demotes information segments in a heterogeneous storage solution based on environmental context, using a graph of interconnected elements, where relevance determines the storage layer, optimizing access by relocating data to higher layers for frequent use and lower layers for infrequent use.
This approach enhances data processing performance by reducing access latency and costs while ensuring accurate and efficient retrieval of relevant data, adapting to user contexts and queries.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
BACKGROUND OF THE INVENTION
[0001] The present disclosure relates to information regions and, more particularly, to using active adaptation to determine relevant segments of an information region.
[0002] For modern data processing systems (particularly information retrieval systems or artificial intelligence systems) to function properly, they require large amounts of stored data. Furthermore, the data must be stored in a way that allows the system to identify and retrieve relevant data at any time. Managing larger data stores increases the likelihood that accurate, appropriate, or satisfactory data will be returned to each input. As data stores grow larger, storage costs can become more significant. This means both the financing costs and the processing costs required to search through ever-larger stores increase. Furthermore, storage solutions capable of managing these large amounts of data are often characterized by slow access, which introduces additional latency into the system.
[0003] US 2018 / 0 218 066 A1 describes a method and system for retrieving information from a plurality of data stores connected to the enterprise data universe. US 2019 / 0 012 373 A1 describes conversational or multiple-question understanding using web intelligence. US 2014 / 0 173 199 A1 describes cluster-centric tiered storage with a flexible tier definition to support transaction performance. US 2011 / 0 302 146 A1 describes technologies for computing a tiering policy that defines how digital objects are desirably stored across a plurality of different storage tiers. SUMMARY
[0004] The invention provides an improved method according to claim 1 and a corresponding improved system and a corresponding improved computer program according to claims 9 and 10. BRIEF DESCRIPTION OF THE DIFFERENT DRAWING VIEWS Fig. 1 illustrates a workflow according to an embodiment illustrated herein for actively determining relevant information segments in an information region. The Fig. 2A and Fig. 2B illustrate techniques for promoting and demoting relevant information segments to improve performance, according to an embodiment disclosed herein. Fig. 3 is a block diagram illustrating an AI system for determining relevant information segments using active adaptation based on environmental contexts, according to an embodiment disclosed herein. Fig. 4 is a flowchart illustrating a method for actively determining relevant information segments according to an embodiment disclosed herein. Fig. 5 is a flowchart illustrating a method for extracting user contexts in an actively adaptive information system, in accordance with an embodiment disclosed herein. Fig. 6 is a flowchart illustrating a method for determining relevant information segments, according to an embodiment disclosed herein. DETAILED DESCRIPTION
[0005] Embodiments of the present disclosure provide techniques for actively promoting and demoting information within an information domain to better discover relevant data efficiently and inexpensively. According to embodiments, information analysis and retrieval systems manage large (and often ever-growing) information domains that can be queried for relevant information. For example, an artificial intelligence (AI) application may utilize such a system to discover relevant data for a given query or input and construct an appropriate response. According to some embodiments of the present disclosure, the information domain is managed as a graph of elements or entities that are interconnected by relationships between those elements.
[0006] According to one embodiment, the system interacts with the information domain, and relationships between entities can be created, removed, or changed (e.g., by increasing or decreasing the weight or significance of the relationship). According to some embodiments, segments and branch domains of the information domain are further annotated based on quantitative observations of the environmental contexts, as well as qualitative feedback from users. According to one embodiment, the information domain is constantly growing as new data is learned and incorporated during processing. According to one embodiment, for example, user inputs and queries can be added to the information domain so that it can later be retrieved and used to better respond to future queries. In this way, the system is dynamic and intelligent. In response to a specific input (e.g.,query from a user), the information system should identify and retrieve relevant data to create a response.
[0007] However, according to embodiments, the information domain can present significant storage challenges. Typically, the prices, capacities, and speeds of storage solutions can vary significantly. For example, serialized storage may be faster than deserialized storage, but at a higher cost (or smaller size). Likewise, random access memory (RAM) or a cache may be relatively faster than disk-based storage, but at a lower capacity and higher price. Furthermore, remote storage solutions (e.g., in the cloud) can provide significant capacity and easy expansion, which in turn comes at a significantly increased latency.According to embodiments of the present disclosure, the continuous storage area represented as a graph is stored in a heterogeneous storage solution comprising a number of layers, each layer being associated with different advantages and disadvantages. According to one embodiment, relatively higher layers are associated with lower access overhead, such as latency during use (e.g., when searching for and retrieving data), while relatively deeper layers are associated with more time or higher overhead during use.
[0008] To provide accurate and reliable answers, according to one embodiment, the system utilizes adaptation to promote and demote segments of the information domain during use so that the most relevant information for a given query is more likely to be found in a relatively higher layer of the heterogeneous storage solution. According to some embodiments, when a segment of the information domain is identified and used based on a specific input, that data is promoted to a relatively higher layer of the storage system. However, as data is used less frequently, it is similarly progressively demoted to lower layers of the system. In this way, relevant data for a specific environmental context is actively relocated within the heterogeneous storage system to improve data processing performance.This means that each time the relevant environmental context reappears, the relevant data is placed in the information space at a relatively higher layer of the system, thus reducing the effort required to access the data. According to embodiments, the environmental context may include a number of system factors such as the current input, previous inputs / queries, the current user, the time of day, and the like.
[0009] According to one embodiment, each query or input represents a question or request to be answered (or otherwise completed) by searching the information space. For example, a user may request information, instruct the performance of a task or operation, and the like. According to one embodiment, in response to queries, the system queries the storage system by first searching higher layers and then (if the query is not yet resolved) lower layers. According to some embodiments, the system decides to proceed to lower (and therefore more costly) storage layers based on a confidence level regarding the current "best possible" answer. For example, the system may retrieve and evaluate data in the highest layer to determine a confidence level that the query is sufficiently resolved.If the confidence level exceeds a predetermined threshold, the data can be used to return an answer. However, if the confidence level falls below a threshold, the system can search the next lower layer of memory, and so on, until a satisfactory answer is found.
[0010] According to some embodiments, the query may also specify a value or importance of the query. According to one embodiment, a cost limit (or low importance) may prevent the storage system from searching the system too deeply. Likewise, a higher cost limit (or greater importance) may cause the system to continue searching in further / deeper layers of the system until a higher confidence answer can be generated. According to one embodiment, the system uses a combination of the strength of relationships between entities and the layer in which each entity is located to determine relevant data in response to a query. For example, the system may determine a first relevant entity (e.g., a node in the graph of the storage system) and further entities (e.g., nodes) that (e.g.,The first node is connected (via an edge in the graph) with a given minimum weight. The system can then retrieve and evaluate the data in each identified element based on the strength of the relationship and / or the layer in which the element is stored.
[0011] According to such an embodiment, the system may, for example, first evaluate data with stronger connections, but wait to evaluate elements stored in deeper layers of the storage system (even if their connection is strong). According to some embodiments, the system holistically determines whether to retrieve data in a deeper layer based on the strength of the connection and the search effort (e.g., for the layer containing the data). According to another embodiment, the system first searches the highest layer and only moves on to the next lower layer after the upper layer has been searched without finding suitable data. In this way, the system can gradually search the graph (reaching increasingly distant nodes and / or increasingly deeper layers of storage) until a satisfactory answer can be returned.
[0012] According to one embodiment, individual user environment contexts can also be profiled. This means that statistical data (e.g., time and / or effort required to generate responses, accuracy and / or confidence of responses, and the like) can be collected for each individual user to determine the most successful profiles (e.g., the best performing and / or most accurate profiles). These can be continuously merged with the less successful profiles to improve the overall performance of the system. According to some embodiments, the most successful profiles are also used to train or guide new users of the system in order to shorten the learning curve and thus the time required for the AI system to adapt to a new user context.
[0013] According to one embodiment, user behavior can, for example, be influenced in a next step by recommendations generated by the system. This means that the system can determine relevant data and / or answer queries partially based on the most successful previously profiled users, thereby increasing performance and reducing time and costs for newer users. According to one embodiment, the profiles consist of a ranking of entities and / or corresponding relationships between entities. These rankings can therefore be easily merged and transferred, thus increasing the overall performance of the system. According to one embodiment, these rankings are learned as the user uses the system.For example, according to some embodiments, the AI system may infer the relationships based on the user's next query, or the user may explicitly indicate the quality of the answer.
[0014] Fig. 1 illustrates a workflow 100 for actively discovering relevant information segments in an information domain, according to an embodiment disclosed herein. As illustrated, a question answering system 105 (e.g., an AI system) receives queries 120 from one or more users and generates corresponding responses 125. According to some embodiments, the queries 120 and the responses 125 are natural language and conversational. According to some embodiments, the queries 120 and the responses 125 are received and returned sequentially, as if the user were conversing with the question answering system 105. For example, an initial query 120 may include "show me the top five commercials sent in the past year," and the corresponding response may include "here are the top five commercials of the past year," along with a list of the commercials.
[0015] According to some embodiments, the system retains one or all previous queries 120 and responses from a particular user as the context of the current query 120 from a user and uses the context to construct the next response 125. Continuing with the above example, after a period of time has elapsed, the user may query "on average, how many of these commercials are opened?" If the question answering system 105 has this context stored, it may construct an accurate response 125 (e.g., "on average, X% of these top commercials were opened in the past year"). According to another embodiment, if the question answering system 105 cannot reliably determine which commercials the user is referring to, the response 125 may request further information (e.g., "Which commercials are you referring to?").
[0016] In the illustrated embodiment, the question answering system 105 uses an information region configured as a graph 115 to generate answers 125. The graph 115, as shown, includes a number of units or elements 110A through H connected by links or edges. According to some embodiments, the links between all elements 110 may be bidirectional or unidirectional, depending on the type of relationship. According to other embodiments, all links are bidirectional. According to one embodiment, the distance between any two elements 110 in the graph 115 is defined based on the number of edges or links that must be traversed on the path from the first element 110 to the second element. According to some embodiments, each edge may have a corresponding weight or strength that indicates the strength of the relationship between the respective elements 110.The graph 115 shown represents a continuous information area, whereby relevant data can be located in each element 110.
[0017] According to some embodiments, the information space grows over the course of interactions between the question answering system 105 and the users (e.g., new elements 110 are created and included, new data is inserted into existing elements 110, and / or connections are made, removed, or modified). According to one embodiment, the graph 115 is stored in a heterogeneous storage system with a number of layers, where each layer has a corresponding access overhead (e.g., in terms of computing resources or time required to access data stored in the layer). A particular element 110 may be stored in any storage layer and may have connections (e.g., pointers) to other elements 110 stored in the same or a different layer.According to one embodiment, elements 110 in the graph 115 are continually relocated over the course of interactions of the question answering system 105 based on their importance to a particular circumstance.
[0018] For example, if element 110G is used to answer a query 120, element 110G can be moved from the layer in which it is stored to a relatively higher layer. That is, the data used to complete query 120 can be stored in a relatively higher layer than the previous layer. When the same (or a similar) context arises again, question answering system 105 can then determine this data much more efficiently and retrieve it with a higher probability.Likewise, according to one embodiment, when the upper layers are full (or when the system determines that a particular element 110 was not useful for a particular context), the question answering system 105 may demote the element 110 to a relatively lower layer to reduce the likelihood that it will be used to answer subsequent queries 120 with similar contexts (and to make room for more relevant data in higher layers).
[0019] The Fig. 2A and Fig. 2B illustrate techniques for upgrading and downgrading to improve performance, according to an embodiment disclosed herein. Fig. In the embodiment illustrated in Figure 2A, the information domain is divided into a series of layers 205A through D. Each layer is characterized by a corresponding access cost. In the present context, a first layer 205 is considered "relatively higher" than a second layer 205 if its access cost is lower than the access cost of the second layer 205. The access cost may include, for example, the latency required to access data stored in layer 205, the computing resources required to access the data, the monetary cost of accessing the data, and the like. Likewise, a first layer 205 is "relatively lower" than a second layer 205 if the access cost for the first layer is higher than the access cost for the second layer 205.
[0020] According to embodiments, the storage system may include any number and type of mechanisms, including volatile memory (e.g., caches or RAM) and / or non-volatile memory (e.g., hard disks). Furthermore, the data within each layer 205 may be stored in any manner. For example, the data may be serialized in one layer 205, but deserialized in another layer 205. Fig. 2A illustrates that items 110 are distributed in information space 200A across layers 205A through D based, at least in part, on how useful or relevant they were in answering previous queries. For example, items 110A and 110B are stored in the highest layer 205A, making them easily accessible to the AI system. This may reduce the access cost for future queries and increase the likelihood that they will be retrieved for a query with similar contexts (e.g., because the system is likely to search at least part of layer 205A before proceeding to the lower layers 205).
[0021] According to embodiments, each layer 205 may have any size. That is, while lower layers 205 tend to be larger and capable of storing greater numbers of elements 110 (e.g., because storage is relatively less expensive financially), according to some embodiments, one or more relatively higher layers 205 may have a higher storage capacity than lower layers 205. For example, while hard disk storage is relatively inexpensive in terms of financial cost per unit of storage, the system may utilize a significant amount of faster (more expensive in terms of financial cost per unit of storage) storage in a higher layer 205 to improve system performance and latency. The particular size of each layer 205 may vary in different implementations.
[0022] Fig. 2A shows an information area 200A at a first time, and Fig. 2B shows the information area 200B at a second point in time. The figures illustrate that in the information area 200B, element 110A has been moved from the highest layer 205A to the relatively lower layer 205B, while element 110C has been moved from the relatively lower layer 205B to the highest layer 205A. Notably, the connections and relationships between elements 110 are unchanged (i.e., the weights have not changed). According to one embodiment, the system may have moved element 110C to the highest layer 205A because it was found to be relevant or useful in answering a current user query. Likewise, element 110A may have been moved to layer 205B because it was either not useful or layer 205A was full.
[0023] Fig. 3 is a block diagram illustrating an AI system 305 according to an embodiment disclosed herein for determining relevant information segments using active adaptation based on environmental contexts. While the AI system 305 is illustrated as a physical computing system, according to embodiments, it may be implemented using hardware or software (e.g., as a virtual computing system) and distributed across any number of devices. According to the illustrated embodiment, the AI system 305 includes a processor 310, a memory 315, a data store 320, and a network interface 325. According to the illustrated embodiment, the processor 310 retrieves and executes program instructions stored in the memory 315 and also stores and retrieves application data stored in the data store 320.Processor 310 is generally representative of a single CPU, multiple CPUs, a single CPU with multiple processor cores, and the like. Likewise, memory 315 is intended to be representative of random access memory. Data storage 320 may be any combination of disk drives, flash memory-based storage devices, and the like, and may include fixed and / or removable storage devices such as hard disk drives, removable memory cards, caches, optical storage, network-attached storage (NAS), or storage area networks (SAN). According to one embodiment, data storage 320 represents a heterogeneous system of layers, each layer corresponding to a different storage type and / or storage with different access costs.The AI system 305 can be connected to one or more other units and components via the network interface 325 to exchange data.
[0024] According to the illustrated embodiment, data store 320 includes data 355 (e.g., items 110 arranged in a graph of information) and one or more profiles 360. According to one embodiment, each profile 360 indicates a ranking of items 110 and / or relationships reflected in data 355 based on their relevance or importance to a particular user and / or context. According to one embodiment, profiles 360 are scored or rated to determine a quality of the profiles based on performance metrics (e.g., the latency or cost for a typical or average response, the confidence associated with a typical response, user feedback on the quality of the responses, and the like). High-quality profiles can then be used to create responses for other users, as discussed in more detail below.
[0025] The figure illustrates that the memory 315 includes an AI application 330, which in turn includes a context component 335, a discovery component 340, an assessment component 345, and a profile component 350. While the functionality of the AI application 330 is depicted as software residing in the main memory 315, it may be implemented using hardware, software, or a combination of hardware and software. Furthermore, the operations of the context component 335, the discovery component 340, the assessment component 345, and the profile component 350 are illustrated as individual components for clarity, but may also be linked to or distributed among any number of components. The AI application 330 generally receives user queries, determines relevant data 355, and creates appropriate responses.According to some embodiments, the AI application 330 further manages the storage of the data 355 by actively promoting (e.g., moving to a relatively higher storage layer) and demoting (e.g., moving to a relatively lower storage layer) data items based on their relevance or usefulness to the current context.
[0026] According to one embodiment, the context component 335 determines the appropriate context for each query. According to one embodiment, the context includes the current query as well as one or more previous queries from the same user. According to some embodiments, the context includes one or more previous responses returned to the user by the AI application 330. According to one embodiment, the context may further include an indication of the user and / or the corresponding profile 360. According to embodiments, the context component 335 may collect and manage this context as the user interacts with the AI application 330 to create better responses for the user. For a given context, the context component 335 may provide the appropriate context to the discovery component 340.
[0027] According to the illustrated embodiment, the discovery component 340 searches the information space (e.g., the data 355) to identify relevant data or information segments based on a particular context. According to one embodiment, the discovery component 340 uses natural language processing (NLP) to parse queries to determine the user's concern and searches the data 355 for relevant information that indicates the expectation. According to embodiments, the discovery component 340 searches based at least in part on the user's current context and / or profile 360. For example, the discovery component 340 may be able to identify more relevant data (or identify relevant data more quickly) using the context.Likewise, the discovery component 340 may use the profile 360 to discover data that is more in line with what the user desires because the profile 360 reflects the user's previous interactions with the AI application 330.
[0028] According to one embodiment, to locate relevant data, the discovery component 340 first searches the highest layer of the data store 320 to locate elements or entities that may be suitable for answering the query. The discovery component 340 may then incrementally discover elements associated with each initially discovered element (e.g., based on connections in the graph structure). According to one embodiment, the discovery component 340 transmits this data to the evaluation component 345 before continuing the search. According to some embodiments, based on the response of the evaluation component 340, the discovery component 340 may continue to discover additional elements in a similar manner (following the connections in the graph) until a satisfactory answer is obtained.
[0029] According to some embodiments, the discovery component 340 retrieves items from the topmost layer until all relevant data is found in the topmost layer, and then proceeds to the deeper layers. According to another embodiment, the discovery component 340 uses the layer identifier as well as the relationship strength to determine whether to retrieve a particular item. According to such an embodiment, for example, an item with a strong connection to a particular item, but located in a deeper layer, may be skipped in favor of a second item in a higher layer, even though the second item has a weaker relationship or lower weight. If the connection is strong enough, the discovery component 340 may equally well retrieve that data if it is located in a deeper layer of the storage 320, according to one embodiment.According to one embodiment, the strength of the relationship between items is based in part on the current context and / or the current 360 profile.
[0030] According to the illustrated embodiment, the evaluation component 345 evaluates the retrieved data items to determine whether the query can be answered using the data. If an answer with sufficiently high confidence can be generated, the evaluation component 345 can instruct the discovery component 340 to terminate the search so that the answer can be generated and returned. If the retrieved data is still insufficient, the evaluation component 345 can instruct the discovery component 340 to continue the search. According to some embodiments, the data in the storage 320 is further ranked and downgraded by the evaluation component 345 based on its relevance. For example, if the confidence of the answer is significantly increased by a particular information segment, the evaluation component 345 can move that information segment to a relatively higher layer.If, however, the confidence level of a particular data element is reduced, the evaluation component 345 can downgrade it to a deeper layer in memory 320. This reduces the probability of being retrieved in similar contexts in the future.
[0031] According to one embodiment, the profile component 350 creates and manages profiles for each user. According to one embodiment, each profile 350 contains a ranking of items and / or relationships between the items, and each profile is compiled by monitoring users' interactions. For example, if two items were used together to create a desired answer by the user, the profile component 350 may update the user profile to indicate that linking the two items would be beneficial (e.g., the weight of the relationship between them should be increased when creating answers for the user). According to some embodiments discussed above, the profile component 350 further determines performance metrics for each profile 360 to determine a quality of the profile 360.
[0032] For example, the profile component 350 may determine how many resources are required to create answers for the user, how accurate or reliable the answers are, how the user rated the answers, and the like. The profile component 350 may then merge higher-quality profiles 360 with lower-quality profiles to improve system performance and guide incoming users (e.g., by selecting potentially relevant data based in part on higher-quality profiles). In this way, the AI application 330 may use the rankings listed in the higher-quality profiles 360 when searching for relevant data 355 and / or when promoting and demoting the data 355 for other users.
[0033] Fig. 4 is a flowchart illustrating a method 400 for actively determining relevant information segments, according to an embodiment disclosed herein. The method 400 begins at block 405, where an AI application 330 receives a query from a user. According to one embodiment, the query includes natural language text (or audio data that can be converted to text). At block 410, the AI application 330 determines the current context of the received query. According to embodiments, this may include an identity of the querying user (or a profile of the querying user), one or more previous queries and / or responses, and the like. The method 400 then proceeds to block 415, where the AI application 330 searches the highest layer of the storage system based on the determined context.The AI application 330 may, for example, use NLP to determine the request and search a graph-based information space to determine data elements relevant to the answer.
[0034] According to the illustrated embodiment, the AI application 330 first searches the highest layer of memory. That is, since the highest layer has the lowest access cost (i.e., it can be searched quickly), the AI application 330 initially limits its search to this layer to reduce the latency required to respond to the user and the required computing resources. The method 400 then proceeds to block 420, where the AI application 330 creates a response based on the retrieved data and calculates a confidence in this created response. According to one embodiment, the confidence may be based on the one or more layers in which the data was found, how well the response matches the identified user concern, and so on. In layer 425, the AI application 330 determines whether the response is satisfactory (e.g.,whether the response meets a predetermined confidence threshold. If so, the method 400 proceeds to block 440, which is discussed in more detail below. Otherwise, the method 400 proceeds to block 430.
[0035] If the answer was still not satisfactory, the AI application 330, in block 430, determines data elements related to the previously determined and retrieved data elements. According to one embodiment, the AI application 330 does this partly based on the relationships defined in the graph structure and partly based on the context of the query (e.g., the user profile, previous queries, and the like). According to embodiments, the determined elements may be located in any layer of the storage system. That is, the previously determined elements may contain pointers to any number of other elements stored in any number of layers in the system.
[0036] The method 400 then proceeds to block 435, where the AI application 330 selects one or more of these recently discovered items to evaluate. According to one embodiment, the AI application 330 determines whether a particular segment of information (e.g., an item in the data) should be retrieved and evaluated based on the relationship strength and the storage layer in which it is located. According to such an embodiment, a stronger relationship is correlated with an increased probability of retrieving the item, and deeper storage layers are correlated with a lower probability of retrieving the item. In this way, the AI application 330 can identify and select data (based on its relationship strength and further based on the fact that it is located in a higher layer of storage where more relevant data is stored) that has a high probability of being more relevant.Then, the method 400 returns to block 420. In this way, the AI application 330 can gradually retrieve and evaluate data, penetrating deeper into the information domain and deeper layers to produce a satisfactory answer while consuming as few resources as possible.
[0037] After a sufficiently satisfactory answer has been created, the AI application 330 returns the created answer to the user in block 440. The method 400 then proceeds to block 445, where the AI application 330 manages the storage system by promoting and / or demoting data items based on the created answer. According to one embodiment, the AI application 330 may, for example, promote (e.g., move to a relatively higher layer) data items that are particularly useful or relevant and demote (e.g., move to a relatively lower layer) data items that were not useful or reduced the confidence of the answer. According to some embodiments, the AI application 330 may receive feedback from the user and use it to further develop its models and operations (e.g., promote and demote data).In this way, the Kl application 330 can better determine relevant data in future work steps.
[0038] Fig. 5 is a flowchart illustrating a method 500 for profiling user contexts in an actively adapting information system, according to an embodiment disclosed herein. The method 500 begins at block 505, where the AI application 330 monitors how users interact with the AI system (e.g., how they ask queries and receive responses). At block 510, the AI application 330 determines, for each user, rankings of entities (e.g., items 110) and / or relationships between entities based on the observed user interaction. The AI application 330 then creates a profile for each user that specifies these rankings. In this way, the AI application 330 can reference the previous user interactions in the form of rankings of the entities and relationships when creating further responses for the user.
[0039] The method 500 then proceeds to block 520, where the AI application 330 determines the most successful profiles. According to embodiments, the AI application 330 may define the best profiles using various key performance indicators (KPIs), such as the average latency of a response, the average data processing cost to create responses, the average accuracy or confidence, and so on. These KPIs are then used to determine the quality of each profile. In block 525, the AI application 330 creates further responses for one or more users based on the determined best profiles. According to one embodiment, the AI application 330, for example, combines the best profiles with the profiles of other users (e.g., new users or users with lower-rated profiles), for example by summarizing or averaging the respective specified rankings.In this way, responses for these users are created partly based on the most successful profiles.
[0040] Fig.6 is a flowchart illustrating a method 600 for determining relevant information segments according to an embodiment disclosed herein. The method 600 begins at block 605, where an AI application 330 receives a first user query. The method 600 then proceeds to block 610, where the AI application 330 retrieves a first information segment from a heterogeneous storage medium, wherein the first information segment is located in an upper layer of the heterogeneous storage medium. The AI application 330 then determines at block 615 that the first information segment does not satisfy the first user query. Based on this determination, the method 600 proceeds to block 620, where the AI application 330 demotes the first information segment to a relatively lower layer of the heterogeneous storage medium. The AI application 330 then retrieves a second information segment from the relatively lower layer of the heterogeneous storage medium at block 625.Then, the method 600 proceeds to block 630 where the AI application 330, after determining that the first user query is satisfied by the second information segment, promotes the second information segment to the higher layer of the heterogeneous storage medium.
[0041] The descriptions of the various embodiments of the present disclosure have been presented for illustrative purposes, but are not intended to be exhaustive or limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terms used herein have been chosen to best explain the principles of the embodiments, the practical application, or technical improvement over commercially available technology, or to enable others skilled in the art to understand the embodiments disclosed herein.
[0042] In the foregoing description, reference is made to the embodiments set forth in this disclosure. However, the scope of the present disclosure is not limited to the individual embodiments described. Rather, it is conceivable that any combination of the foregoing features and elements is capable of implementing and realizing the intended embodiments, regardless of whether these relate to different embodiments. Furthermore, while embodiments disclosed herein may obtain advantages over other possible solutions or over the prior art, the scope of the present disclosure is not limited regardless of whether a particular advantage is obtained by a particular embodiment.Thus, the foregoing aspects, features, embodiments, and advantages are merely illustrative and should not be considered elements or limitations of the appended claims unless expressly relied upon in one or more claims. Likewise, reference to "the invention" should not be construed as a generalization of any inventive subject matter disclosed herein, nor should it be considered elements or limitations of the appended claims unless expressly relied upon in one or more claims.
[0043] Aspects of the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, microcode, etc.), or an embodiment that combines software and hardware aspects, all of which may be generally referred to herein as a "circuit," "module," or "system."
[0044] The present invention may be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to perform aspects of the present invention.
[0045] The computer-readable storage medium may be any physical device capable of retaining and storing instructions for use by an instruction-executing system. The computer-readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. A non-exhaustive list of more specific examples of the computer-readable storage medium includes the following: a portable computer diskette, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM).Flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD), memory stick, floppy disk, mechanically encoded device such as punched cards or raised structures in a groove on which instructions are stored, and any suitable combination thereof. A computer-readable storage medium, as used herein, shall not be construed as carrying transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., pulses of light traveling through an optical fiber cable), or electrical signals carried through a wire.
[0046] Computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to respective computing / processing units or to an external computer or storage unit via a network such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include copper transmission cables, fiber optic transmission lines, wireless transmission, routers, firewalls, switching units, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing unit receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium within the respective computing / processing unit.
[0047] Computer-readable program instructions for performing operations of the present invention may be assembly language instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, or the like, as well as conventional procedural programming languages such as the C programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server.In the latter case, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, over the Internet using an Internet service provider). In some embodiments, electronic circuits, including, for example, programmable logic circuits, field programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), may execute the computer-readable program instructions by utilizing state information of the computer-readable program instructions to personalize the electronic circuits to perform aspects of the present invention.
[0048] Aspects of the present invention are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowcharts and / or block diagrams, as well as combinations of blocks in the flowcharts and / or block diagrams, may be implemented by computer-readable program instructions.
[0049] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that the instructions, executed by the processor of the computer or other programmable data processing apparatus, produce a means for implementing the functions / steps defined in the flowchart block(s) and / or block diagrams.These computer-readable program instructions may also be stored on a computer-readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer-readable storage medium having instructions stored thereon comprises an article of manufacture, including instructions that implement aspects of the function / step specified in the flowchart block(s) and / or block diagrams.
[0050] The computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of process steps to be performed on the computer, other programmable apparatus, or other device to produce a computer-executable process such that the instructions executing on the computer, other programmable apparatus, or other device implement the functions / steps defined in the block(s) of flowcharts and / or block diagrams.
[0051] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowcharts or block diagrams may represent a module, segment, or portion of instructions comprising one or more executable instructions for performing the particular logical function(s). In some alternative implementations, the functions specified in the block may occur in a different order than shown in the figures. For example, two blocks shown in succession may actually execute substantially concurrently, or the blocks may sometimes execute in reverse order depending on the corresponding functionality.It is further understood that each block of the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by special purpose hardware-based systems that perform the specified functions or steps, or by combinations of special purpose hardware and computer instructions.
[0052] Embodiments of the invention may be provided to end users through a cloud computing infrastructure. Cloud computing is generally understood to mean the provision of scalable computing resources as a service over a network. More generally, cloud computing can be defined as a computing capability that provides a separation between the computing resource and its underlying technical architecture (e.g., servers, storage, networks) and enables convenient, on-demand network access to a shared pool of configurable computing resources that can be promptly provisioned and released with minimal administrative effort or interaction from the service provider. Thus, cloud computing provides a user with access to virtual computing resources (e.g.,It enables the deployment of large amounts of data (e.g., storage, data, applications, and even entire virtualized computing systems) in “the cloud,” regardless of the underlying physical systems (or locations of those systems) used to provision the computing resources.
[0053] Typically, cloud computing resources are provided to a user on a pay-per-use basis, where users are only billed for the computing resources actually used (e.g., the amount of storage used by a user or the number of virtualized systems set up by the user). A user can access any resource available in the cloud at any time and from any location on the Internet. In connection with the present invention, a user can access applications (e.g., AI application 330) or corresponding data available in the cloud. For example, AI application 330 may be executed on a computing system in the cloud and operate a heterogeneous storage system.In such a case, the AI application 330 can receive queries and generate responses, and store data elements in various storage layers within the cloud. This enables a user to access this information from any computing system connected to a network connected to the cloud (e.g., the Internet).
Claims
[1] Method (600) comprising: Receiving (605) a first user request (120) by an artificial intelligence (AI) system (305); Retrieving (610) a first information segment (110A) from a heterogeneous storage medium (320), wherein the first information segment is located in an upper layer (205A) of the heterogeneous storage medium; and after determining (615) that the first information segment does not satisfy the first user requirement: downgrading (620) the first information segment to a relatively deeper layer (205B) of the heterogeneous storage medium; Retrieving (625) a second information segment (110C) from the relatively deeper layer of the heterogeneous storage medium; and after determining (630) that the second information segment satisfies the first user request, promoting the second information segment to the upper layer of the heterogeneous storage medium. [2] The method of claim 1, wherein the heterogeneous storage medium comprises a plurality of layers (205A-D), wherein each of the plurality of layers incurs corresponding access costs. [3] The method of claim 2, wherein the upper layer of the heterogeneous storage medium has a relatively lower access cost than the relatively lower layer of the heterogeneous storage medium. [4] The method of claim 1, wherein information segments stored in the relatively lower layer of the heterogeneous storage medium are associated with a relatively lower confidence level than information segments stored in the upper layer of the heterogeneous storage medium. [5] The method of claim 1, wherein the first information segment is determined based on a first context of the first user request, the first context indicating a first user who made the first user request and one or more previous requests of the first user. [6] The method (500, 600) of claim 5, further comprising: Creating (515) a first profile of the first user based on the one or more previous requests of the first user, the first profile indicating a ranking of one or more entities in the first information segment and corresponding relationships between the one or more entities. [7] The method of claim 6, wherein the method further comprises: Determining one or more performance indicators that indicate a quality of the first profile with respect to the AI system; and After determining that the quality of the first profile exceeds a threshold, using the first profile to respond to requests for new users of the AI system. [8] The method of claim 1, wherein the method further comprises: Receiving a second user request by the AI system, the second user request specifying an allowable cost to respond to the second user request; Determining that the second user request cannot be satisfactorily answered using data located in the upper layer of the heterogeneous storage medium; and Search the relatively deeper layer of the heterogeneous storage medium based on the specified allowable cost. [9] A system comprising means suitable for carrying out all the steps of the method according to any preceding claim of the method. [10] A computer program product comprising instructions for carrying out all the steps of the method according to any preceding claim of the method when the computer program is executed on a computer system.
Citation Information
Patent Citations
Determining quality of tier assignments
US20110302146A1
Enhancing Analytics Performance Using Distributed Multi-Tiering
US20140173199A1
Method and system for information retreival
US20180218066A1
Conversational / multi-turn question understanding using web intelligence
US20190012373A1
