METHOD AND SYSTEM FOR PREDICTING SEARCH TYPES FOR THE

INTERNET OF THINGS

The present invention pertains to the technical field of the Internet of things, and more specifically to searching for connected devices in intelligent spaces.

The "Internet of things" means various technical solutions (wireless technologies, Web technologies, computer technologies, TCP/IP, and electronic identification technologies, for example), making it possible to extend the Internet to include the objects that surround us. We can then speak of the Internet of things or, by transposing Web standards, the Web of things.

The Internet of things, interconnecting objects ranging from simple objects identified by electronic labels to devices equipped with actively communicating computer systems, makes it possible to identify these objects, to receive, store, process, and transfer data related to them between/in the physical and virtual environments. These environments are referred to as intelligent spaces.

Here, "intelligent space" means an environment comprising a plurality of sensors, effectors, fluid means of communication (ubiquitous/distributed computing) and on-board intelligence (environmental intelligence) arranged to guarantee information management in a variety of contexts.

As an example of these environments, one may cite:

- A professional environment (an office, a meeting room, an open space, or a company department for example) comprising a plurality of devices such as fixed/mobile telephones, GPSs, desktop/laptop computers, printers, photocopiers, fax machines, cameras, screens, video projectors, external hard disks, or paper files with RFID labels for example;

- A home environment (a bedroom, a house, a hotel, or a residence for example) comprising a plurality of devices such as home appliance systems (alarms, interphones, Smartphones, computers, heating systems, remote controls, televisions, radio receivers, stereo systems, home appliances such as washing machines, freezers, electric cookers,) or manufactured products (labelled lamps, yogurt containers, bottles, or chairs for example); - A medical, industrial, transport, or commercial environment for example, such as an airport, a train station, a shopping centre, a shopping street, a sports complex, or various buildings at a company.

Carried by continuous technological progress (wireless network communication, the development of mobility, ubiquitous computing, the diversification of identification and connection systems) and a profound change in society (autonomous socio-technical environments focused on the individual, personal networks or PANs for Personal Area Networks), the Internet of things is continuously incorporating an increasing number of devices used on a daily basis. The document ("More than 50 Billion connected devices", Ericsson white paper, 284 23-3149 Uen, February 2011), estimates that in the near future more than 50 billion connected devices will be covered by the Internet of things.

Now the problem raised is how to accelerate/improve searching for a specific device from among a very large number of connected devices. In fact, faced with this growing mass of connected devices, the conventional search engines of the Internet of things encounter great difficulty in finding an adequate compromise between accuracy and speed for carrying out a search.

To respond to a search query on the Internet of things, an algorithm browsing a graph of descriptions of connected devices, or a method based on an automatic learning technique, can be adopted to return results above a certain threshold.

However, these search methods can only be implemented when a search context is explicitly indicated in the search query. As a search context is not necessarily known a priori, it is therefore typically difficult to identify the appropriate search type: a search privileging speed or accuracy?

One object of the present invention is to propose a framework that makes it possible to predict a search context on the Internet of things. Another object of the present invention is to predict the type of a search (precise, fuzzy) to be carried out in response to a search query on the Internet of things.

Another object of the present invention is to improve the resolution of search queries on the Internet of things.

Another object of the present invention is to reduce the latency of search engines by predicting a search context.

One object of the present invention is to make a substantial improvement to searching (speed/accuracy) on the Internet of things. Another object of the present invention is to improve search results in a very widely distributed and open information system such as the Internet of things.

For this purpose, the invention relates, according to a first aspect, to a method for predicting the search type for a search query on the Internet of things comprising a step to estimate the distribution of the desired object using a means of calculation.

This method also comprises a step to calculate search type probabilities based on the search types of previous search queries with the same identifier as the search query. This method also comprises a step for weighting the probabilities calculated by weights according to the estimated distributions (that is to say, according to the organisation of an intelligent space: large number of connected devices of various types for example).

The invention relates, according to a second aspect, to a system for predicting the search type in an intelligent space comprising means adapted for studying the distribution of the types of connected devices present in the intelligent space.

This system also comprises a context analyser adapted for extracting, from a search query, a search identifier, and the type of device sought. According to a third aspect, the invention proposes a computer program product implemented on a memory medium, which may be implemented within an information processing unit, and comprises instructions for incrementing the method summarized above. Other characteristics and advantages of the invention will become more clearly and completely apparent upon reading the description below of preferred embodiments, which is done with reference to

Figure 1, which schematically depicts the application context of one embodiment; and

- Figure 2, which schematically depicts a functional, non-limiting depiction of one embodiment.

Figure 1 shows an intelligent space 1 covering a plurality of connected devices 21-37 comprised by the Internet to things. The intelligent space 1 is generally the result of the cooperation and interconnection of a plurality of intelligent sub-spaces 2-4 and/or connected devices 24-26, 32, 37.

A distributed system 5 for information processing (comprising ONSs (Object Naming Services), ubiquitous communication networks, computer hardware/equipment means, and programmed sensors for example) deployed cooperatively in the intelligent space 1 is adapted for supporting the Internet of things in this space. This makes it possible for a user, via a Web interface, to interact (verify the presence, display information, or use a functionality of a connected device, for example) with the connected devices 21-37. For example, this may consist of

- Remotely turning off a lamp equipped with an electronic identification system (an RFID label or a contactless chip for example);

- Starting recording of a film showing on a television channel;

- Verifying the presence of a connected device in a warehouse;

- Recovering information (merchandise transit, validity expiration date, or origin for example) of a connected product. In particular, the distributed information processing system 5 comprises search means (one or more search engines) configured to carry out a search on the Internet of things in response to a search query. Preferably, the intelligent sub-spaces 2-6 are equipped, respectively, with a local search engine, the local search engines being federated by a global search engine.

A search on the Internet of things may also be obtained by implementing distributed solutions that make it possible to support the handling of a large number of search queries.

A search query may be initiated by a user or an application in the space or an intelligent sub-space, through a connected device 21-37 (a computer, a Smartphone, or a PDA for example). When a local search engine receives a search query from a user or an application, a search type prediction system then establishes a search context before carrying out this search.

The "search context" means the search type or mode to be carried out. The main search contexts that can be distinguished are: - A fuzzy search based on the terms of the query or variations of them

(by using for example an edit distance such as the Jaccard distance to define the concept of proximity);

- A precise search based on a precise query syntax.

Figure 2 depicts the various steps that make it possible to predict a search context for each search query received by the search type prediction system. It should be noted that these steps are designed to define a search context and not the search itself, which is the responsibility of the search engine.

To do that, the search type prediction system comprises a context analyser 10.

When received, a search query is analysed by a context analyser 10 to extract

- At least one identifier (query_ID) (arrow 101 in Figure 2). This identifier may be an IP address, a login, a cookie identifier, or any other data that makes it more or less possible to uniquely distinguish the initiator of the query.

- The type of the device sought by this query (arrow 108 in Figure 2).

The search type prediction system checks to see if this query_ID identifier has a search profile in an internal database 11. In other words, the search type prediction system searches to see whether or not the initiator of the query received has a history of queries from the internal database 11.

To that end, the search type prediction system uses the query_ID identifier to recover previous search types corresponding to this identifier from the internal database 11. It should be noted here that the previous search types are recovered, not the entire query/result history associated with the query identifier "query_ID". These data are recovered to calculate statistics on the previous search types (that is to say, created in response to the previous search queries sent by the initiator of the received query), and to calculate search type probabilities from them for the received query.

A test 101 is carried out on the history corresponding to the identifier query_ID found in the internal database 11 to check whether or not it is empty.

If the history is empty (arrow 112 in Figure 2), then the prediction system queries a global discovery module 123. This module 123 comprises a global context analyser 12 that is connected to a plurality of local context analysers 13.

Preferably, the global context analyser 12 queries a local context analyser 13 for an intelligent sub-space that is spatially near the context analyser 10 that received the search query. In other words, if there is no history local to the context analyser 10, the prediction system will use the intelligent sub- spaces located near the initiator of the query.

An intelligent sub-space near the initiator of a search query may be found using information extracted from the received search query (an IP address or GPS coordinates for example). Advantageously, the intelligent sub-spaces close to each other will generally comprise similar connected devices (clusters of devices) constituting a sort of various "interest centres" for the initiator of a query.

The non-empty history recovered from the internal database 11 (arrow 111 in Figure 2) or via the global discovery module 123 (arrow 113 in Figure 2) and corresponding to the identifier query_ID is ranked (step 14 in Figure 2) in a table displaying statistics (total number, number of searches by type) of previous search types as well as their respective probabilities. As an example, this table may comprise the following data: Total searches: 50

Fuzzy searches: 45

Precise searches: 5

Probability(fuzzy searches): 0.9

Probability(precise searches): 0.1 Additionally, the search type prediction system 52 determines the distribution of the type of device sought. A semantic approach may be adopted in particular to identify the type of object requested (arrow 108 in Figure 2) in a search query. Distribution means the organisation of connected devices in an intelligent space: a large number of connected devices, varying types of connected devices for example.

To do this, a connected device analyser 18 makes it possible to study (provide information about) the distribution of the types of connected devices present in the intelligent space 1, such as for example:

- the intelligent space 1 is comprised of a plurality of connected devices of various types or of the same type (a plurality of lamps, a plurality of telephones, for example);

- The connected devices may be spread into several clusters;

- The types of connected devices are highly heterogeneous; The connected device analyser 18 aims to provide a general picture of the connected devices located in the intelligent space 1 in terms of "distribution".

The test 180 of the distribution of the types of connected devices makes it possible to estimate (approach) using means of calculation, the distribution of the type of object sought with a theoretical distribution. As examples, this distribution may be

- A Gaussian distribution 182 (that is to say few clusters of connected devices) that favours (assigns a significant weight to) a precise search over a fuzzy search (step 20 in Figure 2);

- A stochastic distribution 181 (that is to say, a large number of connected devices) that favours (assigns a significant weight to) a fuzzy search over a precise search (step 19 in Figure 2).

The weights assigned to search types (steps 19 and 20) according to the distributions of the type of device sought on the Internet of things are used to refine (step 15 in Figure 2) the search type probabilities calculated in step 14.

As an example, in step 17, the search type probabilities calculated in step 14 are weighted by the factors (weights) assigned to the search types according to the distribution of the type of device sought:

- Probability(a fuzzy search)= 0.9*weight(fuzzy)

- Probability(a precise search) = 0.1*weight(precise)

The prediction system thus calculates a global score (step 17 in Figure 2) that indicates the type of search to be carried out in response to the received search query.

A search type context is obtained using:

- The search type probabilities deduced from previous search types sent by the initiator of the received search query, the previous search types being recovered from the search query history recovered using an identifier extracted from the received search query; and - The distribution of the type of the device requested in the search query among the connected devices.

The search type prediction system weights the search type probabilities, deduced from the search history, using the distributions of connected devices located in the intelligent space. This makes it possible to weight the search history with more or less importance, according to the distribution of devices, for "precise" or "fuzzy" search algorithms.

This makes it possible, in particular, to determine whether a precise search algorithm or a statistical method (fuzzy search, leading to a less precise but more rapid result) should be used to respond to the search query.

Concerning the distribution of devices, four cases can be identified:

- The cardinality of the devices (that is to say the number of connected devices in the intelligent space 1) is low. Therefore, a precise search can be a priority (a search among a low number of devices is certainly rapid);

- The cardinality of devices is high, then to accelerate the search, a statistical approach may be adopted;

- The distribution of devices is Gaussian: some connected devices are of the same device category (that is to say an intelligent space comprising a plurality of lamps, or a plurality of telephones for example). A precise search may be adopted in this case to better distinguish among devices of the same category;

- The distribution of connected devices is stochastic: the objects are of highly varied types. In this case a fuzzy search may be adopted, the probability of distinguishing among devices of the same type thus being low.

Advantageously, the search query does not comprise search context parameters, in particular thresholds.

Predicting the search type to be carried out in searching for connected devices in the Internet of things, according to the method described above, presents several advantages. In particular, it makes it possible to, - Adapt search results to a given context;

- Guarantee the best compromise between search precision and speed on the Internet of things;

- Consider the distribution of connected devices in the type of search to be carried out.